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                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1196         }
1197 }
1198
1199 /*
1200  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1201  */
1202 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1203 {
1204         DEFINE_WAIT(wait);
1205
1206         for (;;) {
1207                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1208
1209                 if (kvm_arch_vcpu_runnable(vcpu)) {
1210                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1211                         break;
1212                 }
1213                 if (kvm_cpu_has_pending_timer(vcpu))
1214                         break;
1215                 if (signal_pending(current))
1216                         break;
1217
1218                 schedule();
1219         }
1220
1221         finish_wait(&vcpu->wq, &wait);
1222 }
1223
1224 void kvm_resched(struct kvm_vcpu *vcpu)
1225 {
1226         if (!need_resched())
1227                 return;
1228         cond_resched();
1229 }
1230 EXPORT_SYMBOL_GPL(kvm_resched);
1231
1232 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1233 {
1234         ktime_t expires;
1235         DEFINE_WAIT(wait);
1236
1237         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1238
1239         /* Sleep for 100 us, and hope lock-holder got scheduled */
1240         expires = ktime_add_ns(ktime_get(), 100000UL);
1241         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1242
1243         finish_wait(&vcpu->wq, &wait);
1244 }
1245 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1246
1247 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1248 {
1249         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1250         struct page *page;
1251
1252         if (vmf->pgoff == 0)
1253                 page = virt_to_page(vcpu->run);
1254 #ifdef CONFIG_X86
1255         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1256                 page = virt_to_page(vcpu->arch.pio_data);
1257 #endif
1258 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1259         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1260                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1261 #endif
1262         else
1263                 return VM_FAULT_SIGBUS;
1264         get_page(page);
1265         vmf->page = page;
1266         return 0;
1267 }
1268
1269 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1270         .fault = kvm_vcpu_fault,
1271 };
1272
1273 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1274 {
1275         vma->vm_ops = &kvm_vcpu_vm_ops;
1276         return 0;
1277 }
1278
1279 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1280 {
1281         struct kvm_vcpu *vcpu = filp->private_data;
1282
1283         kvm_put_kvm(vcpu->kvm);
1284         return 0;
1285 }
1286
1287 static struct file_operations kvm_vcpu_fops = {
1288         .release        = kvm_vcpu_release,
1289         .unlocked_ioctl = kvm_vcpu_ioctl,
1290         .compat_ioctl   = kvm_vcpu_ioctl,
1291         .mmap           = kvm_vcpu_mmap,
1292 };
1293
1294 /*
1295  * Allocates an inode for the vcpu.
1296  */
1297 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1298 {
1299         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1300 }
1301
1302 /*
1303  * Creates some virtual cpus.  Good luck creating more than one.
1304  */
1305 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1306 {
1307         int r;
1308         struct kvm_vcpu *vcpu, *v;
1309
1310         vcpu = kvm_arch_vcpu_create(kvm, id);
1311         if (IS_ERR(vcpu))
1312                 return PTR_ERR(vcpu);
1313
1314         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1315
1316         r = kvm_arch_vcpu_setup(vcpu);
1317         if (r)
1318                 return r;
1319
1320         mutex_lock(&kvm->lock);
1321         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1322                 r = -EINVAL;
1323                 goto vcpu_destroy;
1324         }
1325
1326         kvm_for_each_vcpu(r, v, kvm)
1327                 if (v->vcpu_id == id) {
1328                         r = -EEXIST;
1329                         goto vcpu_destroy;
1330                 }
1331
1332         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1333
1334         /* Now it's all set up, let userspace reach it */
1335         kvm_get_kvm(kvm);
1336         r = create_vcpu_fd(vcpu);
1337         if (r < 0) {
1338                 kvm_put_kvm(kvm);
1339                 goto vcpu_destroy;
1340         }
1341
1342         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1343         smp_wmb();
1344         atomic_inc(&kvm->online_vcpus);
1345
1346 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1347         if (kvm->bsp_vcpu_id == id)
1348                 kvm->bsp_vcpu = vcpu;
1349 #endif
1350         mutex_unlock(&kvm->lock);
1351         return r;
1352
1353 vcpu_destroy:
1354         mutex_unlock(&kvm->lock);
1355         kvm_arch_vcpu_destroy(vcpu);
1356         return r;
1357 }
1358
1359 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1360 {
1361         if (sigset) {
1362                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1363                 vcpu->sigset_active = 1;
1364                 vcpu->sigset = *sigset;
1365         } else
1366                 vcpu->sigset_active = 0;
1367         return 0;
1368 }
1369
1370 static long kvm_vcpu_ioctl(struct file *filp,
1371                            unsigned int ioctl, unsigned long arg)
1372 {
1373         struct kvm_vcpu *vcpu = filp->private_data;
1374         void __user *argp = (void __user *)arg;
1375         int r;
1376         struct kvm_fpu *fpu = NULL;
1377         struct kvm_sregs *kvm_sregs = NULL;
1378
1379         if (vcpu->kvm->mm != current->mm)
1380                 return -EIO;
1381         switch (ioctl) {
1382         case KVM_RUN:
1383                 r = -EINVAL;
1384                 if (arg)
1385                         goto out;
1386                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1387                 break;
1388         case KVM_GET_REGS: {
1389                 struct kvm_regs *kvm_regs;
1390
1391                 r = -ENOMEM;
1392                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1393                 if (!kvm_regs)
1394                         goto out;
1395                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1396                 if (r)
1397                         goto out_free1;
1398                 r = -EFAULT;
1399                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1400                         goto out_free1;
1401                 r = 0;
1402 out_free1:
1403                 kfree(kvm_regs);
1404                 break;
1405         }
1406         case KVM_SET_REGS: {
1407                 struct kvm_regs *kvm_regs;
1408
1409                 r = -ENOMEM;
1410                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1411                 if (!kvm_regs)
1412                         goto out;
1413                 r = -EFAULT;
1414                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1415                         goto out_free2;
1416                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1417                 if (r)
1418                         goto out_free2;
1419                 r = 0;
1420 out_free2:
1421                 kfree(kvm_regs);
1422                 break;
1423         }
1424         case KVM_GET_SREGS: {
1425                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1426                 r = -ENOMEM;
1427                 if (!kvm_sregs)
1428                         goto out;
1429                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1430                 if (r)
1431                         goto out;
1432                 r = -EFAULT;
1433                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1434                         goto out;
1435                 r = 0;
1436                 break;
1437         }
1438         case KVM_SET_SREGS: {
1439                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1440                 r = -ENOMEM;
1441                 if (!kvm_sregs)
1442                         goto out;
1443                 r = -EFAULT;
1444                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1445                         goto out;
1446                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1447                 if (r)
1448                         goto out;
1449                 r = 0;
1450                 break;
1451         }
1452         case KVM_GET_MP_STATE: {
1453                 struct kvm_mp_state mp_state;
1454
1455                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1456                 if (r)
1457                         goto out;
1458                 r = -EFAULT;
1459                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1460                         goto out;
1461                 r = 0;
1462                 break;
1463         }
1464         case KVM_SET_MP_STATE: {
1465                 struct kvm_mp_state mp_state;
1466
1467                 r = -EFAULT;
1468                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1469                         goto out;
1470                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1471                 if (r)
1472                         goto out;
1473                 r = 0;
1474                 break;
1475         }
1476         case KVM_TRANSLATE: {
1477                 struct kvm_translation tr;
1478
1479                 r = -EFAULT;
1480                 if (copy_from_user(&tr, argp, sizeof tr))
1481                         goto out;
1482                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1483                 if (r)
1484                         goto out;
1485                 r = -EFAULT;
1486                 if (copy_to_user(argp, &tr, sizeof tr))
1487                         goto out;
1488                 r = 0;
1489                 break;
1490         }
1491         case KVM_SET_GUEST_DEBUG: {
1492                 struct kvm_guest_debug dbg;
1493
1494                 r = -EFAULT;
1495                 if (copy_from_user(&dbg, argp, sizeof dbg))
1496                         goto out;
1497                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1498                 if (r)
1499                         goto out;
1500                 r = 0;
1501                 break;
1502         }
1503         case KVM_SET_SIGNAL_MASK: {
1504                 struct kvm_signal_mask __user *sigmask_arg = argp;
1505                 struct kvm_signal_mask kvm_sigmask;
1506                 sigset_t sigset, *p;
1507
1508                 p = NULL;
1509                 if (argp) {
1510                         r = -EFAULT;
1511                         if (copy_from_user(&kvm_sigmask, argp,
1512                                            sizeof kvm_sigmask))
1513                                 goto out;
1514                         r = -EINVAL;
1515                         if (kvm_sigmask.len != sizeof sigset)
1516                                 goto out;
1517                         r = -EFAULT;
1518                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1519                                            sizeof sigset))
1520                                 goto out;
1521                         p = &sigset;
1522                 }
1523                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1524                 break;
1525         }
1526         case KVM_GET_FPU: {
1527                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1528                 r = -ENOMEM;
1529                 if (!fpu)
1530                         goto out;
1531                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1532                 if (r)
1533                         goto out;
1534                 r = -EFAULT;
1535                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1536                         goto out;
1537                 r = 0;
1538                 break;
1539         }
1540         case KVM_SET_FPU: {
1541                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1542                 r = -ENOMEM;
1543                 if (!fpu)
1544                         goto out;
1545                 r = -EFAULT;
1546                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1547                         goto out;
1548                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1549                 if (r)
1550                         goto out;
1551                 r = 0;
1552                 break;
1553         }
1554         default:
1555                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1556         }
1557 out:
1558         kfree(fpu);
1559         kfree(kvm_sregs);
1560         return r;
1561 }
1562
1563 static long kvm_vm_ioctl(struct file *filp,
1564                            unsigned int ioctl, unsigned long arg)
1565 {
1566         struct kvm *kvm = filp->private_data;
1567         void __user *argp = (void __user *)arg;
1568         int r;
1569
1570         if (kvm->mm != current->mm)
1571                 return -EIO;
1572         switch (ioctl) {
1573         case KVM_CREATE_VCPU:
1574                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1575                 if (r < 0)
1576                         goto out;
1577                 break;
1578         case KVM_SET_USER_MEMORY_REGION: {
1579                 struct kvm_userspace_memory_region kvm_userspace_mem;
1580
1581                 r = -EFAULT;
1582                 if (copy_from_user(&kvm_userspace_mem, argp,
1583                                                 sizeof kvm_userspace_mem))
1584                         goto out;
1585
1586                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1587                 if (r)
1588                         goto out;
1589                 break;
1590         }
1591         case KVM_GET_DIRTY_LOG: {
1592                 struct kvm_dirty_log log;
1593
1594                 r = -EFAULT;
1595                 if (copy_from_user(&log, argp, sizeof log))
1596                         goto out;
1597                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1598                 if (r)
1599                         goto out;
1600                 break;
1601         }
1602 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1603         case KVM_REGISTER_COALESCED_MMIO: {
1604                 struct kvm_coalesced_mmio_zone zone;
1605                 r = -EFAULT;
1606                 if (copy_from_user(&zone, argp, sizeof zone))
1607                         goto out;
1608                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1609                 if (r)
1610                         goto out;
1611                 r = 0;
1612                 break;
1613         }
1614         case KVM_UNREGISTER_COALESCED_MMIO: {
1615                 struct kvm_coalesced_mmio_zone zone;
1616                 r = -EFAULT;
1617                 if (copy_from_user(&zone, argp, sizeof zone))
1618                         goto out;
1619                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1620                 if (r)
1621                         goto out;
1622                 r = 0;
1623                 break;
1624         }
1625 #endif
1626         case KVM_IRQFD: {
1627                 struct kvm_irqfd data;
1628
1629                 r = -EFAULT;
1630                 if (copy_from_user(&data, argp, sizeof data))
1631                         goto out;
1632                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1633                 break;
1634         }
1635         case KVM_IOEVENTFD: {
1636                 struct kvm_ioeventfd data;
1637
1638                 r = -EFAULT;
1639                 if (copy_from_user(&data, argp, sizeof data))
1640                         goto out;
1641                 r = kvm_ioeventfd(kvm, &data);
1642                 break;
1643         }
1644 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1645         case KVM_SET_BOOT_CPU_ID:
1646                 r = 0;
1647                 mutex_lock(&kvm->lock);
1648                 if (atomic_read(&kvm->online_vcpus) != 0)
1649                         r = -EBUSY;
1650                 else
1651                         kvm->bsp_vcpu_id = arg;
1652                 mutex_unlock(&kvm->lock);
1653                 break;
1654 #endif
1655         default:
1656                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1657                 if (r == -ENOTTY)
1658                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1659         }
1660 out:
1661         return r;
1662 }
1663
1664 #ifdef CONFIG_COMPAT
1665 struct compat_kvm_dirty_log {
1666         __u32 slot;
1667         __u32 padding1;
1668         union {
1669                 compat_uptr_t dirty_bitmap; /* one bit per page */
1670                 __u64 padding2;
1671         };
1672 };
1673
1674 static long kvm_vm_compat_ioctl(struct file *filp,
1675                            unsigned int ioctl, unsigned long arg)
1676 {
1677         struct kvm *kvm = filp->private_data;
1678         int r;
1679
1680         if (kvm->mm != current->mm)
1681                 return -EIO;
1682         switch (ioctl) {
1683         case KVM_GET_DIRTY_LOG: {
1684                 struct compat_kvm_dirty_log compat_log;
1685                 struct kvm_dirty_log log;
1686
1687                 r = -EFAULT;
1688                 if (copy_from_user(&compat_log, (void __user *)arg,
1689                                    sizeof(compat_log)))
1690                         goto out;
1691                 log.slot         = compat_log.slot;
1692                 log.padding1     = compat_log.padding1;
1693                 log.padding2     = compat_log.padding2;
1694                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1695
1696                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1697                 if (r)
1698                         goto out;
1699                 break;
1700         }
1701         default:
1702                 r = kvm_vm_ioctl(filp, ioctl, arg);
1703         }
1704
1705 out:
1706         return r;
1707 }
1708 #endif
1709
1710 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1711 {
1712         struct page *page[1];
1713         unsigned long addr;
1714         int npages;
1715         gfn_t gfn = vmf->pgoff;
1716         struct kvm *kvm = vma->vm_file->private_data;
1717
1718         addr = gfn_to_hva(kvm, gfn);
1719         if (kvm_is_error_hva(addr))
1720                 return VM_FAULT_SIGBUS;
1721
1722         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1723                                 NULL);
1724         if (unlikely(npages != 1))
1725                 return VM_FAULT_SIGBUS;
1726
1727         vmf->page = page[0];
1728         return 0;
1729 }
1730
1731 static const struct vm_operations_struct kvm_vm_vm_ops = {
1732         .fault = kvm_vm_fault,
1733 };
1734
1735 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1736 {
1737         vma->vm_ops = &kvm_vm_vm_ops;
1738         return 0;
1739 }
1740
1741 static struct file_operations kvm_vm_fops = {
1742         .release        = kvm_vm_release,
1743         .unlocked_ioctl = kvm_vm_ioctl,
1744 #ifdef CONFIG_COMPAT
1745         .compat_ioctl   = kvm_vm_compat_ioctl,
1746 #endif
1747         .mmap           = kvm_vm_mmap,
1748 };
1749
1750 static int kvm_dev_ioctl_create_vm(void)
1751 {
1752         int fd, r;
1753         struct kvm *kvm;
1754
1755         kvm = kvm_create_vm();
1756         if (IS_ERR(kvm))
1757                 return PTR_ERR(kvm);
1758 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1759         r = kvm_coalesced_mmio_init(kvm);
1760         if (r < 0) {
1761                 kvm_put_kvm(kvm);
1762                 return r;
1763         }
1764 #endif
1765         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1766         if (fd < 0)
1767                 kvm_put_kvm(kvm);
1768
1769         return fd;
1770 }
1771
1772 static long kvm_dev_ioctl_check_extension_generic(long arg)
1773 {
1774         switch (arg) {
1775         case KVM_CAP_USER_MEMORY:
1776         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1777         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1778 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1779         case KVM_CAP_SET_BOOT_CPU_ID:
1780 #endif
1781         case KVM_CAP_INTERNAL_ERROR_DATA:
1782                 return 1;
1783 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1784         case KVM_CAP_IRQ_ROUTING:
1785                 return KVM_MAX_IRQ_ROUTES;
1786 #endif
1787         default:
1788                 break;
1789         }
1790         return kvm_dev_ioctl_check_extension(arg);
1791 }
1792
1793 static long kvm_dev_ioctl(struct file *filp,
1794                           unsigned int ioctl, unsigned long arg)
1795 {
1796         long r = -EINVAL;
1797
1798         switch (ioctl) {
1799         case KVM_GET_API_VERSION:
1800                 r = -EINVAL;
1801                 if (arg)
1802                         goto out;
1803                 r = KVM_API_VERSION;
1804                 break;
1805         case KVM_CREATE_VM:
1806                 r = -EINVAL;
1807                 if (arg)
1808                         goto out;
1809                 r = kvm_dev_ioctl_create_vm();
1810                 break;
1811         case KVM_CHECK_EXTENSION:
1812                 r = kvm_dev_ioctl_check_extension_generic(arg);
1813                 break;
1814         case KVM_GET_VCPU_MMAP_SIZE:
1815                 r = -EINVAL;
1816                 if (arg)
1817                         goto out;
1818                 r = PAGE_SIZE;     /* struct kvm_run */
1819 #ifdef CONFIG_X86
1820                 r += PAGE_SIZE;    /* pio data page */
1821 #endif
1822 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1823                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1824 #endif
1825                 break;
1826         case KVM_TRACE_ENABLE:
1827         case KVM_TRACE_PAUSE:
1828         case KVM_TRACE_DISABLE:
1829                 r = -EOPNOTSUPP;
1830                 break;
1831         default:
1832                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1833         }
1834 out:
1835         return r;
1836 }
1837
1838 static struct file_operations kvm_chardev_ops = {
1839         .unlocked_ioctl = kvm_dev_ioctl,
1840         .compat_ioctl   = kvm_dev_ioctl,
1841 };
1842
1843 static struct miscdevice kvm_dev = {
1844         KVM_MINOR,
1845         "kvm",
1846         &kvm_chardev_ops,
1847 };
1848
1849 static void hardware_enable(void *junk)
1850 {
1851         int cpu = raw_smp_processor_id();
1852         int r;
1853
1854         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1855                 return;
1856
1857         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1858
1859         r = kvm_arch_hardware_enable(NULL);
1860
1861         if (r) {
1862                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1863                 atomic_inc(&hardware_enable_failed);
1864                 printk(KERN_INFO "kvm: enabling virtualization on "
1865                                  "CPU%d failed\n", cpu);
1866         }
1867 }
1868
1869 static void hardware_disable(void *junk)
1870 {
1871         int cpu = raw_smp_processor_id();
1872
1873         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1874                 return;
1875         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1876         kvm_arch_hardware_disable(NULL);
1877 }
1878
1879 static void hardware_disable_all_nolock(void)
1880 {
1881         BUG_ON(!kvm_usage_count);
1882
1883         kvm_usage_count--;
1884         if (!kvm_usage_count)
1885                 on_each_cpu(hardware_disable, NULL, 1);
1886 }
1887
1888 static void hardware_disable_all(void)
1889 {
1890         spin_lock(&kvm_lock);
1891         hardware_disable_all_nolock();
1892         spin_unlock(&kvm_lock);
1893 }
1894
1895 static int hardware_enable_all(void)
1896 {
1897         int r = 0;
1898
1899         spin_lock(&kvm_lock);
1900
1901         kvm_usage_count++;
1902         if (kvm_usage_count == 1) {
1903                 atomic_set(&hardware_enable_failed, 0);
1904                 on_each_cpu(hardware_enable, NULL, 1);
1905
1906                 if (atomic_read(&hardware_enable_failed)) {
1907                         hardware_disable_all_nolock();
1908                         r = -EBUSY;
1909                 }
1910         }
1911
1912         spin_unlock(&kvm_lock);
1913
1914         return r;
1915 }
1916
1917 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1918                            void *v)
1919 {
1920         int cpu = (long)v;
1921
1922         if (!kvm_usage_count)
1923                 return NOTIFY_OK;
1924
1925         val &= ~CPU_TASKS_FROZEN;
1926         switch (val) {
1927         case CPU_DYING:
1928                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1929                        cpu);
1930                 hardware_disable(NULL);
1931                 break;
1932         case CPU_ONLINE:
1933                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1934                        cpu);
1935                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1936                 break;
1937         }
1938         return NOTIFY_OK;
1939 }
1940
1941
1942 asmlinkage void kvm_handle_fault_on_reboot(void)
1943 {
1944         if (kvm_rebooting)
1945                 /* spin while reset goes on */
1946                 while (true)
1947                         ;
1948         /* Fault while not rebooting.  We want the trace. */
1949         BUG();
1950 }
1951 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1952
1953 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1954                       void *v)
1955 {
1956         /*
1957          * Some (well, at least mine) BIOSes hang on reboot if
1958          * in vmx root mode.
1959          *
1960          * And Intel TXT required VMX off for all cpu when system shutdown.
1961          */
1962         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1963         kvm_rebooting = true;
1964         on_each_cpu(hardware_disable, NULL, 1);
1965         return NOTIFY_OK;
1966 }
1967
1968 static struct notifier_block kvm_reboot_notifier = {
1969         .notifier_call = kvm_reboot,
1970         .priority = 0,
1971 };
1972
1973 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1974 {
1975         int i;
1976
1977         for (i = 0; i < bus->dev_count; i++) {
1978                 struct kvm_io_device *pos = bus->devs[i];
1979
1980                 kvm_iodevice_destructor(pos);
1981         }
1982         kfree(bus);
1983 }
1984
1985 /* kvm_io_bus_write - called under kvm->slots_lock */
1986 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1987                      int len, const void *val)
1988 {
1989         int i;
1990         struct kvm_io_bus *bus;
1991
1992         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
1993         for (i = 0; i < bus->dev_count; i++)
1994                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1995                         return 0;
1996         return -EOPNOTSUPP;
1997 }
1998
1999 /* kvm_io_bus_read - called under kvm->slots_lock */
2000 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2001                     int len, void *val)
2002 {
2003         int i;
2004         struct kvm_io_bus *bus;
2005
2006         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2007         for (i = 0; i < bus->dev_count; i++)
2008                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2009                         return 0;
2010         return -EOPNOTSUPP;
2011 }
2012
2013 /* Caller must hold slots_lock. */
2014 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2015                             struct kvm_io_device *dev)
2016 {
2017         struct kvm_io_bus *new_bus, *bus;
2018
2019         bus = kvm->buses[bus_idx];
2020         if (bus->dev_count > NR_IOBUS_DEVS-1)
2021                 return -ENOSPC;
2022
2023         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2024         if (!new_bus)
2025                 return -ENOMEM;
2026         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2027         new_bus->devs[new_bus->dev_count++] = dev;
2028         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2029         synchronize_srcu_expedited(&kvm->srcu);
2030         kfree(bus);
2031
2032         return 0;
2033 }
2034
2035 /* Caller must hold slots_lock. */
2036 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2037                               struct kvm_io_device *dev)
2038 {
2039         int i, r;
2040         struct kvm_io_bus *new_bus, *bus;
2041
2042         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2043         if (!new_bus)
2044                 return -ENOMEM;
2045
2046         bus = kvm->buses[bus_idx];
2047         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2048
2049         r = -ENOENT;
2050         for (i = 0; i < new_bus->dev_count; i++)
2051                 if (new_bus->devs[i] == dev) {
2052                         r = 0;
2053                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2054                         break;
2055                 }
2056
2057         if (r) {
2058                 kfree(new_bus);
2059                 return r;
2060         }
2061
2062         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2063         synchronize_srcu_expedited(&kvm->srcu);
2064         kfree(bus);
2065         return r;
2066 }
2067
2068 static struct notifier_block kvm_cpu_notifier = {
2069         .notifier_call = kvm_cpu_hotplug,
2070         .priority = 20, /* must be > scheduler priority */
2071 };
2072
2073 static int vm_stat_get(void *_offset, u64 *val)
2074 {
2075         unsigned offset = (long)_offset;
2076         struct kvm *kvm;
2077
2078         *val = 0;
2079         spin_lock(&kvm_lock);
2080         list_for_each_entry(kvm, &vm_list, vm_list)
2081                 *val += *(u32 *)((void *)kvm + offset);
2082         spin_unlock(&kvm_lock);
2083         return 0;
2084 }
2085
2086 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2087
2088 static int vcpu_stat_get(void *_offset, u64 *val)
2089 {
2090         unsigned offset = (long)_offset;
2091         struct kvm *kvm;
2092         struct kvm_vcpu *vcpu;
2093         int i;
2094
2095         *val = 0;
2096         spin_lock(&kvm_lock);
2097         list_for_each_entry(kvm, &vm_list, vm_list)
2098                 kvm_for_each_vcpu(i, vcpu, kvm)
2099                         *val += *(u32 *)((void *)vcpu + offset);
2100
2101         spin_unlock(&kvm_lock);
2102         return 0;
2103 }
2104
2105 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2106
2107 static const struct file_operations *stat_fops[] = {
2108         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2109         [KVM_STAT_VM]   = &vm_stat_fops,
2110 };
2111
2112 static void kvm_init_debug(void)
2113 {
2114         struct kvm_stats_debugfs_item *p;
2115
2116         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2117         for (p = debugfs_entries; p->name; ++p)
2118                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2119                                                 (void *)(long)p->offset,
2120                                                 stat_fops[p->kind]);
2121 }
2122
2123 static void kvm_exit_debug(void)
2124 {
2125         struct kvm_stats_debugfs_item *p;
2126
2127         for (p = debugfs_entries; p->name; ++p)
2128                 debugfs_remove(p->dentry);
2129         debugfs_remove(kvm_debugfs_dir);
2130 }
2131
2132 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2133 {
2134         if (kvm_usage_count)
2135                 hardware_disable(NULL);
2136         return 0;
2137 }
2138
2139 static int kvm_resume(struct sys_device *dev)
2140 {
2141         if (kvm_usage_count)
2142                 hardware_enable(NULL);
2143         return 0;
2144 }
2145
2146 static struct sysdev_class kvm_sysdev_class = {
2147         .name = "kvm",
2148         .suspend = kvm_suspend,
2149         .resume = kvm_resume,
2150 };
2151
2152 static struct sys_device kvm_sysdev = {
2153         .id = 0,
2154         .cls = &kvm_sysdev_class,
2155 };
2156
2157 struct page *bad_page;
2158 pfn_t bad_pfn;
2159
2160 static inline
2161 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2162 {
2163         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2164 }
2165
2166 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2167 {
2168         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2169
2170         kvm_arch_vcpu_load(vcpu, cpu);
2171 }
2172
2173 static void kvm_sched_out(struct preempt_notifier *pn,
2174                           struct task_struct *next)
2175 {
2176         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2177
2178         kvm_arch_vcpu_put(vcpu);
2179 }
2180
2181 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2182                   struct module *module)
2183 {
2184         int r;
2185         int cpu;
2186
2187         r = kvm_arch_init(opaque);
2188         if (r)
2189                 goto out_fail;
2190
2191         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2192
2193         if (bad_page == NULL) {
2194                 r = -ENOMEM;
2195                 goto out;
2196         }
2197
2198         bad_pfn = page_to_pfn(bad_page);
2199
2200         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2201                 r = -ENOMEM;
2202                 goto out_free_0;
2203         }
2204
2205         r = kvm_arch_hardware_setup();
2206         if (r < 0)
2207                 goto out_free_0a;
2208
2209         for_each_online_cpu(cpu) {
2210                 smp_call_function_single(cpu,
2211                                 kvm_arch_check_processor_compat,
2212                                 &r, 1);
2213                 if (r < 0)
2214                         goto out_free_1;
2215         }
2216
2217         r = register_cpu_notifier(&kvm_cpu_notifier);
2218         if (r)
2219                 goto out_free_2;
2220         register_reboot_notifier(&kvm_reboot_notifier);
2221
2222         r = sysdev_class_register(&kvm_sysdev_class);
2223         if (r)
2224                 goto out_free_3;
2225
2226         r = sysdev_register(&kvm_sysdev);
2227         if (r)
2228                 goto out_free_4;
2229
2230         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2231         if (!vcpu_align)
2232                 vcpu_align = __alignof__(struct kvm_vcpu);
2233         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2234                                            0, NULL);
2235         if (!kvm_vcpu_cache) {
2236                 r = -ENOMEM;
2237                 goto out_free_5;
2238         }
2239
2240         kvm_chardev_ops.owner = module;
2241         kvm_vm_fops.owner = module;
2242         kvm_vcpu_fops.owner = module;
2243
2244         r = misc_register(&kvm_dev);
2245         if (r) {
2246                 printk(KERN_ERR "kvm: misc device register failed\n");
2247                 goto out_free;
2248         }
2249
2250         kvm_preempt_ops.sched_in = kvm_sched_in;
2251         kvm_preempt_ops.sched_out = kvm_sched_out;
2252
2253         kvm_init_debug();
2254
2255         return 0;
2256
2257 out_free:
2258         kmem_cache_destroy(kvm_vcpu_cache);
2259 out_free_5:
2260         sysdev_unregister(&kvm_sysdev);
2261 out_free_4:
2262         sysdev_class_unregister(&kvm_sysdev_class);
2263 out_free_3:
2264         unregister_reboot_notifier(&kvm_reboot_notifier);
2265         unregister_cpu_notifier(&kvm_cpu_notifier);
2266 out_free_2:
2267 out_free_1:
2268         kvm_arch_hardware_unsetup();
2269 out_free_0a:
2270         free_cpumask_var(cpus_hardware_enabled);
2271 out_free_0:
2272         __free_page(bad_page);
2273 out:
2274         kvm_arch_exit();
2275 out_fail:
2276         return r;
2277 }
2278 EXPORT_SYMBOL_GPL(kvm_init);
2279
2280 void kvm_exit(void)
2281 {
2282         kvm_exit_debug();
2283         misc_deregister(&kvm_dev);
2284         kmem_cache_destroy(kvm_vcpu_cache);
2285         sysdev_unregister(&kvm_sysdev);
2286         sysdev_class_unregister(&kvm_sysdev_class);
2287         unregister_reboot_notifier(&kvm_reboot_notifier);
2288         unregister_cpu_notifier(&kvm_cpu_notifier);
2289         on_each_cpu(hardware_disable, NULL, 1);
2290         kvm_arch_hardware_unsetup();
2291         kvm_arch_exit();
2292         free_cpumask_var(cpus_hardware_enabled);
2293         __free_page(bad_page);
2294 }
2295 EXPORT_SYMBOL_GPL(kvm_exit);