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         if (!npages)
 561                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 562
 563         new = old = *memslot;
 564
 565         new.base_gfn = base_gfn;
 566         new.npages = npages;
 567         new.flags = mem->flags;
 568
 569         /* Disallow changing a memory slot's size. */
 570         r = -EINVAL;
 571         if (npages && old.npages && npages != old.npages)
 572                 goto out_free;
 573
 574         /* Check for overlaps */
 575         r = -EEXIST;
 576         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 577                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 578
 579                 if (s == memslot || !s->npages)
 580                         continue;
 581                 if (!((base_gfn + npages <= s->base_gfn) ||
 582                       (base_gfn >= s->base_gfn + s->npages)))
 583                         goto out_free;
 584         }
 585
 586         /* Free page dirty bitmap if unneeded */
 587         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 588                 new.dirty_bitmap = NULL;
 589
 590         r = -ENOMEM;
 591
 592         /* Allocate if a slot is being created */
 593 #ifndef CONFIG_S390
 594         if (npages && !new.rmap) {
 595                 new.rmap = vmalloc(npages * sizeof(struct page *));
 596
 597                 if (!new.rmap)
 598                         goto out_free;
 599
 600                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 601
 602                 new.user_alloc = user_alloc;
 603                 new.userspace_addr = mem->userspace_addr;
 604         }
 605         if (!npages)
 606                 goto skip_lpage;
 607
 608         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 609                 unsigned long ugfn;
 610                 unsigned long j;
 611                 int lpages;
 612                 int level = i + 2;
 613
 614                 /* Avoid unused variable warning if no large pages */
 615                 (void)level;
 616
 617                 if (new.lpage_info[i])
 618                         continue;
 619
 620                 lpages = 1 + (base_gfn + npages - 1) /
 621                              KVM_PAGES_PER_HPAGE(level);
 622                 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
 623
 624                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 625
 626                 if (!new.lpage_info[i])
 627                         goto out_free;
 628
 629                 memset(new.lpage_info[i], 0,
 630                        lpages * sizeof(*new.lpage_info[i]));
 631
 632                 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
 633                         new.lpage_info[i][0].write_count = 1;
 634                 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
 635                         new.lpage_info[i][lpages - 1].write_count = 1;
 636                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 637                 /*
 638                  * If the gfn and userspace address are not aligned wrt each
 639                  * other, or if explicitly asked to, disable large page
 640                  * support for this slot
 641                  */
 642                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 643                     !largepages_enabled)
 644                         for (j = 0; j < lpages; ++j)
 645                                 new.lpage_info[i][j].write_count = 1;
 646         }
 647
 648 skip_lpage:
 649
 650         /* Allocate page dirty bitmap if needed */
 651         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 652                 unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(&new);
 653
 654                 new.dirty_bitmap = vmalloc(dirty_bytes);
 655                 if (!new.dirty_bitmap)
 656                         goto out_free;
 657                 memset(new.dirty_bitmap, 0, dirty_bytes);
 658                 /* destroy any largepage mappings for dirty tracking */
 659                 if (old.npages)
 660                         flush_shadow = 1;
 661         }
 662 #else  /* not defined CONFIG_S390 */
 663         new.user_alloc = user_alloc;
 664         if (user_alloc)
 665                 new.userspace_addr = mem->userspace_addr;
 666 #endif /* not defined CONFIG_S390 */
 667
 668         if (!npages) {
 669                 r = -ENOMEM;
 670                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 671                 if (!slots)
 672                         goto out_free;
 673                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 674                 if (mem->slot >= slots->nmemslots)
 675                         slots->nmemslots = mem->slot + 1;
 676                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 677
 678                 old_memslots = kvm->memslots;
 679                 rcu_assign_pointer(kvm->memslots, slots);
 680                 synchronize_srcu_expedited(&kvm->srcu);
 681                 /* From this point no new shadow pages pointing to a deleted
 682                  * memslot will be created.
 683                  *
 684                  * validation of sp->gfn happens in:
 685                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 686                  *      - kvm_is_visible_gfn (mmu_check_roots)
 687                  */
 688                 kvm_arch_flush_shadow(kvm);
 689                 kfree(old_memslots);
 690         }
 691
 692         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 693         if (r)
 694                 goto out_free;
 695
 696 #ifdef CONFIG_DMAR
 697         /* map the pages in iommu page table */
 698         if (npages) {
 699                 r = kvm_iommu_map_pages(kvm, &new);
 700                 if (r)
 701                         goto out_free;
 702         }
 703 #endif
 704
 705         r = -ENOMEM;
 706         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 707         if (!slots)
 708                 goto out_free;
 709         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 710         if (mem->slot >= slots->nmemslots)
 711                 slots->nmemslots = mem->slot + 1;
 712
 713         /* actual memory is freed via old in kvm_free_physmem_slot below */
 714         if (!npages) {
 715                 new.rmap = NULL;
 716                 new.dirty_bitmap = NULL;
 717                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 718                         new.lpage_info[i] = NULL;
 719         }
 720
 721         slots->memslots[mem->slot] = new;
 722         old_memslots = kvm->memslots;
 723         rcu_assign_pointer(kvm->memslots, slots);
 724         synchronize_srcu_expedited(&kvm->srcu);
 725
 726         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 727
 728         kvm_free_physmem_slot(&old, &new);
 729         kfree(old_memslots);
 730
 731         if (flush_shadow)
 732                 kvm_arch_flush_shadow(kvm);
 733
 734         return 0;
 735
 736 out_free:
 737         kvm_free_physmem_slot(&new, &old);
 738 out:
 739         return r;
 740
 741 }
 742 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 743
 744 int kvm_set_memory_region(struct kvm *kvm,
 745                           struct kvm_userspace_memory_region *mem,
 746                           int user_alloc)
 747 {
 748         int r;
 749
 750         mutex_lock(&kvm->slots_lock);
 751         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 752         mutex_unlock(&kvm->slots_lock);
 753         return r;
 754 }
 755 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 756
 757 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 758                                    struct
 759                                    kvm_userspace_memory_region *mem,
 760                                    int user_alloc)
 761 {
 762         if (mem->slot >= KVM_MEMORY_SLOTS)
 763                 return -EINVAL;
 764         return kvm_set_memory_region(kvm, mem, user_alloc);
 765 }
 766
 767 int kvm_get_dirty_log(struct kvm *kvm,
 768                         struct kvm_dirty_log *log, int *is_dirty)
 769 {
 770         struct kvm_memory_slot *memslot;
 771         int r, i;
 772         unsigned long n;
 773         unsigned long any = 0;
 774
 775         r = -EINVAL;
 776         if (log->slot >= KVM_MEMORY_SLOTS)
 777                 goto out;
 778
 779         memslot = &kvm->memslots->memslots[log->slot];
 780         r = -ENOENT;
 781         if (!memslot->dirty_bitmap)
 782                 goto out;
 783
 784         n = kvm_dirty_bitmap_bytes(memslot);
 785
 786         for (i = 0; !any && i < n/sizeof(long); ++i)
 787                 any = memslot->dirty_bitmap[i];
 788
 789         r = -EFAULT;
 790         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 791                 goto out;
 792
 793         if (any)
 794                 *is_dirty = 1;
 795
 796         r = 0;
 797 out:
 798         return r;
 799 }
 800
 801 void kvm_disable_largepages(void)
 802 {
 803         largepages_enabled = false;
 804 }
 805 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 806
 807 int is_error_page(struct page *page)
 808 {
 809         return page == bad_page;
 810 }
 811 EXPORT_SYMBOL_GPL(is_error_page);
 812
 813 int is_error_pfn(pfn_t pfn)
 814 {
 815         return pfn == bad_pfn;
 816 }
 817 EXPORT_SYMBOL_GPL(is_error_pfn);
 818
 819 static inline unsigned long bad_hva(void)
 820 {
 821         return PAGE_OFFSET;
 822 }
 823
 824 int kvm_is_error_hva(unsigned long addr)
 825 {
 826         return addr == bad_hva();
 827 }
 828 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 829
 830 struct kvm_memory_slot *gfn_to_memslot_unaliased(struct kvm *kvm, gfn_t gfn)
 831 {
 832         int i;
 833         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 834
 835         for (i = 0; i < slots->nmemslots; ++i) {
 836                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 837
 838                 if (gfn >= memslot->base_gfn
 839                     && gfn < memslot->base_gfn + memslot->npages)
 840                         return memslot;
 841         }
 842         return NULL;
 843 }
 844 EXPORT_SYMBOL_GPL(gfn_to_memslot_unaliased);
 845
 846 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 847 {
 848         gfn = unalias_gfn(kvm, gfn);
 849         return gfn_to_memslot_unaliased(kvm, gfn);
 850 }
 851
 852 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 853 {
 854         int i;
 855         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 856
 857         gfn = unalias_gfn_instantiation(kvm, gfn);
 858         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 859                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 860
 861                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 862                         continue;
 863
 864                 if (gfn >= memslot->base_gfn
 865                     && gfn < memslot->base_gfn + memslot->npages)
 866                         return 1;
 867         }
 868         return 0;
 869 }
 870 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 871
 872 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 873 {
 874         struct vm_area_struct *vma;
 875         unsigned long addr, size;
 876
 877         size = PAGE_SIZE;
 878
 879         addr = gfn_to_hva(kvm, gfn);
 880         if (kvm_is_error_hva(addr))
 881                 return PAGE_SIZE;
 882
 883         down_read(&current->mm->mmap_sem);
 884         vma = find_vma(current->mm, addr);
 885         if (!vma)
 886                 goto out;
 887
 888         size = vma_kernel_pagesize(vma);
 889
 890 out:
 891         up_read(&current->mm->mmap_sem);
 892
 893         return size;
 894 }
 895
 896 int memslot_id(struct kvm *kvm, gfn_t gfn)
 897 {
 898         int i;
 899         struct kvm_memslots *slots = rcu_dereference(kvm->memslots);
 900         struct kvm_memory_slot *memslot = NULL;
 901
 902         gfn = unalias_gfn(kvm, gfn);
 903         for (i = 0; i < slots->nmemslots; ++i) {
 904                 memslot = &slots->memslots[i];
 905
 906                 if (gfn >= memslot->base_gfn
 907                     && gfn < memslot->base_gfn + memslot->npages)
 908                         break;
 909         }
 910
 911         return memslot - slots->memslots;
 912 }
 913
 914 static unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
 915 {
 916         return slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE;
 917 }
 918
 919 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 920 {
 921         struct kvm_memory_slot *slot;
 922
 923         gfn = unalias_gfn_instantiation(kvm, gfn);
 924         slot = gfn_to_memslot_unaliased(kvm, gfn);
 925         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 926                 return bad_hva();
 927         return gfn_to_hva_memslot(slot, gfn);
 928 }
 929 EXPORT_SYMBOL_GPL(gfn_to_hva);
 930
 931 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr)
 932 {
 933         struct page *page[1];
 934         int npages;
 935         pfn_t pfn;
 936
 937         might_sleep();
 938
 939         npages = get_user_pages_fast(addr, 1, 1, page);
 940
 941         if (unlikely(npages != 1)) {
 942                 struct vm_area_struct *vma;
 943
 944                 down_read(&current->mm->mmap_sem);
 945                 vma = find_vma(current->mm, addr);
 946
 947                 if (vma == NULL || addr < vma->vm_start ||
 948                     !(vma->vm_flags & VM_PFNMAP)) {
 949                         up_read(&current->mm->mmap_sem);
 950                         get_page(bad_page);
 951                         return page_to_pfn(bad_page);
 952                 }
 953
 954                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
 955                 up_read(&current->mm->mmap_sem);
 956                 BUG_ON(!kvm_is_mmio_pfn(pfn));
 957         } else
 958                 pfn = page_to_pfn(page[0]);
 959
 960         return pfn;
 961 }
 962
 963 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
 964 {
 965         unsigned long addr;
 966
 967         addr = gfn_to_hva(kvm, gfn);
 968         if (kvm_is_error_hva(addr)) {
 969                 get_page(bad_page);
 970                 return page_to_pfn(bad_page);
 971         }
 972
 973         return hva_to_pfn(kvm, addr);
 974 }
 975 EXPORT_SYMBOL_GPL(gfn_to_pfn);
 976
 977 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
 978                          struct kvm_memory_slot *slot, gfn_t gfn)
 979 {
 980         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
 981         return hva_to_pfn(kvm, addr);
 982 }
 983
 984 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
 985 {
 986         pfn_t pfn;
 987
 988         pfn = gfn_to_pfn(kvm, gfn);
 989         if (!kvm_is_mmio_pfn(pfn))
 990                 return pfn_to_page(pfn);
 991
 992         WARN_ON(kvm_is_mmio_pfn(pfn));
 993
 994         get_page(bad_page);
 995         return bad_page;
 996 }
 997
 998 EXPORT_SYMBOL_GPL(gfn_to_page);
 999
1000 void kvm_release_page_clean(struct page *page)
1001 {
1002         kvm_release_pfn_clean(page_to_pfn(page));
1003 }
1004 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1005
1006 void kvm_release_pfn_clean(pfn_t pfn)
1007 {
1008         if (!kvm_is_mmio_pfn(pfn))
1009                 put_page(pfn_to_page(pfn));
1010 }
1011 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1012
1013 void kvm_release_page_dirty(struct page *page)
1014 {
1015         kvm_release_pfn_dirty(page_to_pfn(page));
1016 }
1017 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1018
1019 void kvm_release_pfn_dirty(pfn_t pfn)
1020 {
1021         kvm_set_pfn_dirty(pfn);
1022         kvm_release_pfn_clean(pfn);
1023 }
1024 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1025
1026 void kvm_set_page_dirty(struct page *page)
1027 {
1028         kvm_set_pfn_dirty(page_to_pfn(page));
1029 }
1030 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1031
1032 void kvm_set_pfn_dirty(pfn_t pfn)
1033 {
1034         if (!kvm_is_mmio_pfn(pfn)) {
1035                 struct page *page = pfn_to_page(pfn);
1036                 if (!PageReserved(page))
1037                         SetPageDirty(page);
1038         }
1039 }
1040 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1041
1042 void kvm_set_pfn_accessed(pfn_t pfn)
1043 {
1044         if (!kvm_is_mmio_pfn(pfn))
1045                 mark_page_accessed(pfn_to_page(pfn));
1046 }
1047 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1048
1049 void kvm_get_pfn(pfn_t pfn)
1050 {
1051         if (!kvm_is_mmio_pfn(pfn))
1052                 get_page(pfn_to_page(pfn));
1053 }
1054 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1055
1056 static int next_segment(unsigned long len, int offset)
1057 {
1058         if (len > PAGE_SIZE - offset)
1059                 return PAGE_SIZE - offset;
1060         else
1061                 return len;
1062 }
1063
1064 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1065                         int len)
1066 {
1067         int r;
1068         unsigned long addr;
1069
1070         addr = gfn_to_hva(kvm, gfn);
1071         if (kvm_is_error_hva(addr))
1072                 return -EFAULT;
1073         r = copy_from_user(data, (void __user *)addr + offset, len);
1074         if (r)
1075                 return -EFAULT;
1076         return 0;
1077 }
1078 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1079
1080 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1081 {
1082         gfn_t gfn = gpa >> PAGE_SHIFT;
1083         int seg;
1084         int offset = offset_in_page(gpa);
1085         int ret;
1086
1087         while ((seg = next_segment(len, offset)) != 0) {
1088                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1089                 if (ret < 0)
1090                         return ret;
1091                 offset = 0;
1092                 len -= seg;
1093                 data += seg;
1094                 ++gfn;
1095         }
1096         return 0;
1097 }
1098 EXPORT_SYMBOL_GPL(kvm_read_guest);
1099
1100 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1101                           unsigned long len)
1102 {
1103         int r;
1104         unsigned long addr;
1105         gfn_t gfn = gpa >> PAGE_SHIFT;
1106         int offset = offset_in_page(gpa);
1107
1108         addr = gfn_to_hva(kvm, gfn);
1109         if (kvm_is_error_hva(addr))
1110                 return -EFAULT;
1111         pagefault_disable();
1112         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1113         pagefault_enable();
1114         if (r)
1115                 return -EFAULT;
1116         return 0;
1117 }
1118 EXPORT_SYMBOL(kvm_read_guest_atomic);
1119
1120 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1121                          int offset, int len)
1122 {
1123         int r;
1124         unsigned long addr;
1125
1126         addr = gfn_to_hva(kvm, gfn);
1127         if (kvm_is_error_hva(addr))
1128                 return -EFAULT;
1129         r = copy_to_user((void __user *)addr + offset, data, len);
1130         if (r)
1131                 return -EFAULT;
1132         mark_page_dirty(kvm, gfn);
1133         return 0;
1134 }
1135 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1136
1137 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1138                     unsigned long len)
1139 {
1140         gfn_t gfn = gpa >> PAGE_SHIFT;
1141         int seg;
1142         int offset = offset_in_page(gpa);
1143         int ret;
1144
1145         while ((seg = next_segment(len, offset)) != 0) {
1146                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1147                 if (ret < 0)
1148                         return ret;
1149                 offset = 0;
1150                 len -= seg;
1151                 data += seg;
1152                 ++gfn;
1153         }
1154         return 0;
1155 }
1156
1157 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1158 {
1159         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1160 }
1161 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1162
1163 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1164 {
1165         gfn_t gfn = gpa >> PAGE_SHIFT;
1166         int seg;
1167         int offset = offset_in_page(gpa);
1168         int ret;
1169
1170         while ((seg = next_segment(len, offset)) != 0) {
1171                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1172                 if (ret < 0)
1173                         return ret;
1174                 offset = 0;
1175                 len -= seg;
1176                 ++gfn;
1177         }
1178         return 0;
1179 }
1180 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1181
1182 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1183 {
1184         struct kvm_memory_slot *memslot;
1185
1186         gfn = unalias_gfn(kvm, gfn);
1187         memslot = gfn_to_memslot_unaliased(kvm, gfn);
1188         if (memslot && memslot->dirty_bitmap) {
1189                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1190                 unsigned long *p = memslot->dirty_bitmap +
1191                                         rel_gfn / BITS_PER_LONG;
1192                 int offset = rel_gfn % BITS_PER_LONG;
1193
1194                 /* avoid RMW */
1195                 if (!generic_test_le_bit(offset, p))
1196                         generic___set_le_bit(offset, p);
1197         }
1198 }
1199
1200 /*
1201  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1202  */
1203 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1204 {
1205         DEFINE_WAIT(wait);
1206
1207         for (;;) {
1208                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1209
1210                 if (kvm_arch_vcpu_runnable(vcpu)) {
1211                         set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1212                         break;
1213                 }
1214                 if (kvm_cpu_has_pending_timer(vcpu))
1215                         break;
1216                 if (signal_pending(current))
1217                         break;
1218
1219                 schedule();
1220         }
1221
1222         finish_wait(&vcpu->wq, &wait);
1223 }
1224
1225 void kvm_resched(struct kvm_vcpu *vcpu)
1226 {
1227         if (!need_resched())
1228                 return;
1229         cond_resched();
1230 }
1231 EXPORT_SYMBOL_GPL(kvm_resched);
1232
1233 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1234 {
1235         ktime_t expires;
1236         DEFINE_WAIT(wait);
1237
1238         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1239
1240         /* Sleep for 100 us, and hope lock-holder got scheduled */
1241         expires = ktime_add_ns(ktime_get(), 100000UL);
1242         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1243
1244         finish_wait(&vcpu->wq, &wait);
1245 }
1246 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1247
1248 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1249 {
1250         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1251         struct page *page;
1252
1253         if (vmf->pgoff == 0)
1254                 page = virt_to_page(vcpu->run);
1255 #ifdef CONFIG_X86
1256         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1257                 page = virt_to_page(vcpu->arch.pio_data);
1258 #endif
1259 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1260         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1261                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1262 #endif
1263         else
1264                 return VM_FAULT_SIGBUS;
1265         get_page(page);
1266         vmf->page = page;
1267         return 0;
1268 }
1269
1270 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1271         .fault = kvm_vcpu_fault,
1272 };
1273
1274 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1275 {
1276         vma->vm_ops = &kvm_vcpu_vm_ops;
1277         return 0;
1278 }
1279
1280 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1281 {
1282         struct kvm_vcpu *vcpu = filp->private_data;
1283
1284         kvm_put_kvm(vcpu->kvm);
1285         return 0;
1286 }
1287
1288 static struct file_operations kvm_vcpu_fops = {
1289         .release        = kvm_vcpu_release,
1290         .unlocked_ioctl = kvm_vcpu_ioctl,
1291         .compat_ioctl   = kvm_vcpu_ioctl,
1292         .mmap           = kvm_vcpu_mmap,
1293 };
1294
1295 /*
1296  * Allocates an inode for the vcpu.
1297  */
1298 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1299 {
1300         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1301 }
1302
1303 /*
1304  * Creates some virtual cpus.  Good luck creating more than one.
1305  */
1306 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1307 {
1308         int r;
1309         struct kvm_vcpu *vcpu, *v;
1310
1311         vcpu = kvm_arch_vcpu_create(kvm, id);
1312         if (IS_ERR(vcpu))
1313                 return PTR_ERR(vcpu);
1314
1315         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1316
1317         r = kvm_arch_vcpu_setup(vcpu);
1318         if (r)
1319                 return r;
1320
1321         mutex_lock(&kvm->lock);
1322         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1323                 r = -EINVAL;
1324                 goto vcpu_destroy;
1325         }
1326
1327         kvm_for_each_vcpu(r, v, kvm)
1328                 if (v->vcpu_id == id) {
1329                         r = -EEXIST;
1330                         goto vcpu_destroy;
1331                 }
1332
1333         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1334
1335         /* Now it's all set up, let userspace reach it */
1336         kvm_get_kvm(kvm);
1337         r = create_vcpu_fd(vcpu);
1338         if (r < 0) {
1339                 kvm_put_kvm(kvm);
1340                 goto vcpu_destroy;
1341         }
1342
1343         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1344         smp_wmb();
1345         atomic_inc(&kvm->online_vcpus);
1346
1347 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1348         if (kvm->bsp_vcpu_id == id)
1349                 kvm->bsp_vcpu = vcpu;
1350 #endif
1351         mutex_unlock(&kvm->lock);
1352         return r;
1353
1354 vcpu_destroy:
1355         mutex_unlock(&kvm->lock);
1356         kvm_arch_vcpu_destroy(vcpu);
1357         return r;
1358 }
1359
1360 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1361 {
1362         if (sigset) {
1363                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1364                 vcpu->sigset_active = 1;
1365                 vcpu->sigset = *sigset;
1366         } else
1367                 vcpu->sigset_active = 0;
1368         return 0;
1369 }
1370
1371 static long kvm_vcpu_ioctl(struct file *filp,
1372                            unsigned int ioctl, unsigned long arg)
1373 {
1374         struct kvm_vcpu *vcpu = filp->private_data;
1375         void __user *argp = (void __user *)arg;
1376         int r;
1377         struct kvm_fpu *fpu = NULL;
1378         struct kvm_sregs *kvm_sregs = NULL;
1379
1380         if (vcpu->kvm->mm != current->mm)
1381                 return -EIO;
1382         switch (ioctl) {
1383         case KVM_RUN:
1384                 r = -EINVAL;
1385                 if (arg)
1386                         goto out;
1387                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1388                 break;
1389         case KVM_GET_REGS: {
1390                 struct kvm_regs *kvm_regs;
1391
1392                 r = -ENOMEM;
1393                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1394                 if (!kvm_regs)
1395                         goto out;
1396                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1397                 if (r)
1398                         goto out_free1;
1399                 r = -EFAULT;
1400                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1401                         goto out_free1;
1402                 r = 0;
1403 out_free1:
1404                 kfree(kvm_regs);
1405                 break;
1406         }
1407         case KVM_SET_REGS: {
1408                 struct kvm_regs *kvm_regs;
1409
1410                 r = -ENOMEM;
1411                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1412                 if (!kvm_regs)
1413                         goto out;
1414                 r = -EFAULT;
1415                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1416                         goto out_free2;
1417                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1418                 if (r)
1419                         goto out_free2;
1420                 r = 0;
1421 out_free2:
1422                 kfree(kvm_regs);
1423                 break;
1424         }
1425         case KVM_GET_SREGS: {
1426                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1427                 r = -ENOMEM;
1428                 if (!kvm_sregs)
1429                         goto out;
1430                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1431                 if (r)
1432                         goto out;
1433                 r = -EFAULT;
1434                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1435                         goto out;
1436                 r = 0;
1437                 break;
1438         }
1439         case KVM_SET_SREGS: {
1440                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1441                 r = -ENOMEM;
1442                 if (!kvm_sregs)
1443                         goto out;
1444                 r = -EFAULT;
1445                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1446                         goto out;
1447                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1448                 if (r)
1449                         goto out;
1450                 r = 0;
1451                 break;
1452         }
1453         case KVM_GET_MP_STATE: {
1454                 struct kvm_mp_state mp_state;
1455
1456                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1457                 if (r)
1458                         goto out;
1459                 r = -EFAULT;
1460                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1461                         goto out;
1462                 r = 0;
1463                 break;
1464         }
1465         case KVM_SET_MP_STATE: {
1466                 struct kvm_mp_state mp_state;
1467
1468                 r = -EFAULT;
1469                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1470                         goto out;
1471                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1472                 if (r)
1473                         goto out;
1474                 r = 0;
1475                 break;
1476         }
1477         case KVM_TRANSLATE: {
1478                 struct kvm_translation tr;
1479
1480                 r = -EFAULT;
1481                 if (copy_from_user(&tr, argp, sizeof tr))
1482                         goto out;
1483                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1484                 if (r)
1485                         goto out;
1486                 r = -EFAULT;
1487                 if (copy_to_user(argp, &tr, sizeof tr))
1488                         goto out;
1489                 r = 0;
1490                 break;
1491         }
1492         case KVM_SET_GUEST_DEBUG: {
1493                 struct kvm_guest_debug dbg;
1494
1495                 r = -EFAULT;
1496                 if (copy_from_user(&dbg, argp, sizeof dbg))
1497                         goto out;
1498                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1499                 if (r)
1500                         goto out;
1501                 r = 0;
1502                 break;
1503         }
1504         case KVM_SET_SIGNAL_MASK: {
1505                 struct kvm_signal_mask __user *sigmask_arg = argp;
1506                 struct kvm_signal_mask kvm_sigmask;
1507                 sigset_t sigset, *p;
1508
1509                 p = NULL;
1510                 if (argp) {
1511                         r = -EFAULT;
1512                         if (copy_from_user(&kvm_sigmask, argp,
1513                                            sizeof kvm_sigmask))
1514                                 goto out;
1515                         r = -EINVAL;
1516                         if (kvm_sigmask.len != sizeof sigset)
1517                                 goto out;
1518                         r = -EFAULT;
1519                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1520                                            sizeof sigset))
1521                                 goto out;
1522                         p = &sigset;
1523                 }
1524                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1525                 break;
1526         }
1527         case KVM_GET_FPU: {
1528                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1529                 r = -ENOMEM;
1530                 if (!fpu)
1531                         goto out;
1532                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1533                 if (r)
1534                         goto out;
1535                 r = -EFAULT;
1536                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1537                         goto out;
1538                 r = 0;
1539                 break;
1540         }
1541         case KVM_SET_FPU: {
1542                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1543                 r = -ENOMEM;
1544                 if (!fpu)
1545                         goto out;
1546                 r = -EFAULT;
1547                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1548                         goto out;
1549                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1550                 if (r)
1551                         goto out;
1552                 r = 0;
1553                 break;
1554         }
1555         default:
1556                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1557         }
1558 out:
1559         kfree(fpu);
1560         kfree(kvm_sregs);
1561         return r;
1562 }
1563
1564 static long kvm_vm_ioctl(struct file *filp,
1565                            unsigned int ioctl, unsigned long arg)
1566 {
1567         struct kvm *kvm = filp->private_data;
1568         void __user *argp = (void __user *)arg;
1569         int r;
1570
1571         if (kvm->mm != current->mm)
1572                 return -EIO;
1573         switch (ioctl) {
1574         case KVM_CREATE_VCPU:
1575                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1576                 if (r < 0)
1577                         goto out;
1578                 break;
1579         case KVM_SET_USER_MEMORY_REGION: {
1580                 struct kvm_userspace_memory_region kvm_userspace_mem;
1581
1582                 r = -EFAULT;
1583                 if (copy_from_user(&kvm_userspace_mem, argp,
1584                                                 sizeof kvm_userspace_mem))
1585                         goto out;
1586
1587                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1588                 if (r)
1589                         goto out;
1590                 break;
1591         }
1592         case KVM_GET_DIRTY_LOG: {
1593                 struct kvm_dirty_log log;
1594
1595                 r = -EFAULT;
1596                 if (copy_from_user(&log, argp, sizeof log))
1597                         goto out;
1598                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1599                 if (r)
1600                         goto out;
1601                 break;
1602         }
1603 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1604         case KVM_REGISTER_COALESCED_MMIO: {
1605                 struct kvm_coalesced_mmio_zone zone;
1606                 r = -EFAULT;
1607                 if (copy_from_user(&zone, argp, sizeof zone))
1608                         goto out;
1609                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1610                 if (r)
1611                         goto out;
1612                 r = 0;
1613                 break;
1614         }
1615         case KVM_UNREGISTER_COALESCED_MMIO: {
1616                 struct kvm_coalesced_mmio_zone zone;
1617                 r = -EFAULT;
1618                 if (copy_from_user(&zone, argp, sizeof zone))
1619                         goto out;
1620                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1621                 if (r)
1622                         goto out;
1623                 r = 0;
1624                 break;
1625         }
1626 #endif
1627         case KVM_IRQFD: {
1628                 struct kvm_irqfd data;
1629
1630                 r = -EFAULT;
1631                 if (copy_from_user(&data, argp, sizeof data))
1632                         goto out;
1633                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1634                 break;
1635         }
1636         case KVM_IOEVENTFD: {
1637                 struct kvm_ioeventfd data;
1638
1639                 r = -EFAULT;
1640                 if (copy_from_user(&data, argp, sizeof data))
1641                         goto out;
1642                 r = kvm_ioeventfd(kvm, &data);
1643                 break;
1644         }
1645 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1646         case KVM_SET_BOOT_CPU_ID:
1647                 r = 0;
1648                 mutex_lock(&kvm->lock);
1649                 if (atomic_read(&kvm->online_vcpus) != 0)
1650                         r = -EBUSY;
1651                 else
1652                         kvm->bsp_vcpu_id = arg;
1653                 mutex_unlock(&kvm->lock);
1654                 break;
1655 #endif
1656         default:
1657                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1658                 if (r == -ENOTTY)
1659                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1660         }
1661 out:
1662         return r;
1663 }
1664
1665 #ifdef CONFIG_COMPAT
1666 struct compat_kvm_dirty_log {
1667         __u32 slot;
1668         __u32 padding1;
1669         union {
1670                 compat_uptr_t dirty_bitmap; /* one bit per page */
1671                 __u64 padding2;
1672         };
1673 };
1674
1675 static long kvm_vm_compat_ioctl(struct file *filp,
1676                            unsigned int ioctl, unsigned long arg)
1677 {
1678         struct kvm *kvm = filp->private_data;
1679         int r;
1680
1681         if (kvm->mm != current->mm)
1682                 return -EIO;
1683         switch (ioctl) {
1684         case KVM_GET_DIRTY_LOG: {
1685                 struct compat_kvm_dirty_log compat_log;
1686                 struct kvm_dirty_log log;
1687
1688                 r = -EFAULT;
1689                 if (copy_from_user(&compat_log, (void __user *)arg,
1690                                    sizeof(compat_log)))
1691                         goto out;
1692                 log.slot         = compat_log.slot;
1693                 log.padding1     = compat_log.padding1;
1694                 log.padding2     = compat_log.padding2;
1695                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1696
1697                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1698                 if (r)
1699                         goto out;
1700                 break;
1701         }
1702         default:
1703                 r = kvm_vm_ioctl(filp, ioctl, arg);
1704         }
1705
1706 out:
1707         return r;
1708 }
1709 #endif
1710
1711 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1712 {
1713         struct page *page[1];
1714         unsigned long addr;
1715         int npages;
1716         gfn_t gfn = vmf->pgoff;
1717         struct kvm *kvm = vma->vm_file->private_data;
1718
1719         addr = gfn_to_hva(kvm, gfn);
1720         if (kvm_is_error_hva(addr))
1721                 return VM_FAULT_SIGBUS;
1722
1723         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1724                                 NULL);
1725         if (unlikely(npages != 1))
1726                 return VM_FAULT_SIGBUS;
1727
1728         vmf->page = page[0];
1729         return 0;
1730 }
1731
1732 static const struct vm_operations_struct kvm_vm_vm_ops = {
1733         .fault = kvm_vm_fault,
1734 };
1735
1736 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1737 {
1738         vma->vm_ops = &kvm_vm_vm_ops;
1739         return 0;
1740 }
1741
1742 static struct file_operations kvm_vm_fops = {
1743         .release        = kvm_vm_release,
1744         .unlocked_ioctl = kvm_vm_ioctl,
1745 #ifdef CONFIG_COMPAT
1746         .compat_ioctl   = kvm_vm_compat_ioctl,
1747 #endif
1748         .mmap           = kvm_vm_mmap,
1749 };
1750
1751 static int kvm_dev_ioctl_create_vm(void)
1752 {
1753         int fd, r;
1754         struct kvm *kvm;
1755
1756         kvm = kvm_create_vm();
1757         if (IS_ERR(kvm))
1758                 return PTR_ERR(kvm);
1759 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1760         r = kvm_coalesced_mmio_init(kvm);
1761         if (r < 0) {
1762                 kvm_put_kvm(kvm);
1763                 return r;
1764         }
1765 #endif
1766         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1767         if (fd < 0)
1768                 kvm_put_kvm(kvm);
1769
1770         return fd;
1771 }
1772
1773 static long kvm_dev_ioctl_check_extension_generic(long arg)
1774 {
1775         switch (arg) {
1776         case KVM_CAP_USER_MEMORY:
1777         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1778         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1779 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1780         case KVM_CAP_SET_BOOT_CPU_ID:
1781 #endif
1782         case KVM_CAP_INTERNAL_ERROR_DATA:
1783                 return 1;
1784 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1785         case KVM_CAP_IRQ_ROUTING:
1786                 return KVM_MAX_IRQ_ROUTES;
1787 #endif
1788         default:
1789                 break;
1790         }
1791         return kvm_dev_ioctl_check_extension(arg);
1792 }
1793
1794 static long kvm_dev_ioctl(struct file *filp,
1795                           unsigned int ioctl, unsigned long arg)
1796 {
1797         long r = -EINVAL;
1798
1799         switch (ioctl) {
1800         case KVM_GET_API_VERSION:
1801                 r = -EINVAL;
1802                 if (arg)
1803                         goto out;
1804                 r = KVM_API_VERSION;
1805                 break;
1806         case KVM_CREATE_VM:
1807                 r = -EINVAL;
1808                 if (arg)
1809                         goto out;
1810                 r = kvm_dev_ioctl_create_vm();
1811                 break;
1812         case KVM_CHECK_EXTENSION:
1813                 r = kvm_dev_ioctl_check_extension_generic(arg);
1814                 break;
1815         case KVM_GET_VCPU_MMAP_SIZE:
1816                 r = -EINVAL;
1817                 if (arg)
1818                         goto out;
1819                 r = PAGE_SIZE;     /* struct kvm_run */
1820 #ifdef CONFIG_X86
1821                 r += PAGE_SIZE;    /* pio data page */
1822 #endif
1823 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1824                 r += PAGE_SIZE;    /* coalesced mmio ring page */
1825 #endif
1826                 break;
1827         case KVM_TRACE_ENABLE:
1828         case KVM_TRACE_PAUSE:
1829         case KVM_TRACE_DISABLE:
1830                 r = -EOPNOTSUPP;
1831                 break;
1832         default:
1833                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1834         }
1835 out:
1836         return r;
1837 }
1838
1839 static struct file_operations kvm_chardev_ops = {
1840         .unlocked_ioctl = kvm_dev_ioctl,
1841         .compat_ioctl   = kvm_dev_ioctl,
1842 };
1843
1844 static struct miscdevice kvm_dev = {
1845         KVM_MINOR,
1846         "kvm",
1847         &kvm_chardev_ops,
1848 };
1849
1850 static void hardware_enable(void *junk)
1851 {
1852         int cpu = raw_smp_processor_id();
1853         int r;
1854
1855         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
1856                 return;
1857
1858         cpumask_set_cpu(cpu, cpus_hardware_enabled);
1859
1860         r = kvm_arch_hardware_enable(NULL);
1861
1862         if (r) {
1863                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1864                 atomic_inc(&hardware_enable_failed);
1865                 printk(KERN_INFO "kvm: enabling virtualization on "
1866                                  "CPU%d failed\n", cpu);
1867         }
1868 }
1869
1870 static void hardware_disable(void *junk)
1871 {
1872         int cpu = raw_smp_processor_id();
1873
1874         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
1875                 return;
1876         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1877         kvm_arch_hardware_disable(NULL);
1878 }
1879
1880 static void hardware_disable_all_nolock(void)
1881 {
1882         BUG_ON(!kvm_usage_count);
1883
1884         kvm_usage_count--;
1885         if (!kvm_usage_count)
1886                 on_each_cpu(hardware_disable, NULL, 1);
1887 }
1888
1889 static void hardware_disable_all(void)
1890 {
1891         spin_lock(&kvm_lock);
1892         hardware_disable_all_nolock();
1893         spin_unlock(&kvm_lock);
1894 }
1895
1896 static int hardware_enable_all(void)
1897 {
1898         int r = 0;
1899
1900         spin_lock(&kvm_lock);
1901
1902         kvm_usage_count++;
1903         if (kvm_usage_count == 1) {
1904                 atomic_set(&hardware_enable_failed, 0);
1905                 on_each_cpu(hardware_enable, NULL, 1);
1906
1907                 if (atomic_read(&hardware_enable_failed)) {
1908                         hardware_disable_all_nolock();
1909                         r = -EBUSY;
1910                 }
1911         }
1912
1913         spin_unlock(&kvm_lock);
1914
1915         return r;
1916 }
1917
1918 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1919                            void *v)
1920 {
1921         int cpu = (long)v;
1922
1923         if (!kvm_usage_count)
1924                 return NOTIFY_OK;
1925
1926         val &= ~CPU_TASKS_FROZEN;
1927         switch (val) {
1928         case CPU_DYING:
1929                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1930                        cpu);
1931                 hardware_disable(NULL);
1932                 break;
1933         case CPU_UP_CANCELED:
1934                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1935                        cpu);
1936                 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1937                 break;
1938         case CPU_ONLINE:
1939                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1940                        cpu);
1941                 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1942                 break;
1943         }
1944         return NOTIFY_OK;
1945 }
1946
1947
1948 asmlinkage void kvm_handle_fault_on_reboot(void)
1949 {
1950         if (kvm_rebooting)
1951                 /* spin while reset goes on */
1952                 while (true)
1953                         ;
1954         /* Fault while not rebooting.  We want the trace. */
1955         BUG();
1956 }
1957 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1958
1959 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1960                       void *v)
1961 {
1962         /*
1963          * Some (well, at least mine) BIOSes hang on reboot if
1964          * in vmx root mode.
1965          *
1966          * And Intel TXT required VMX off for all cpu when system shutdown.
1967          */
1968         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1969         kvm_rebooting = true;
1970         on_each_cpu(hardware_disable, NULL, 1);
1971         return NOTIFY_OK;
1972 }
1973
1974 static struct notifier_block kvm_reboot_notifier = {
1975         .notifier_call = kvm_reboot,
1976         .priority = 0,
1977 };
1978
1979 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1980 {
1981         int i;
1982
1983         for (i = 0; i < bus->dev_count; i++) {
1984                 struct kvm_io_device *pos = bus->devs[i];
1985
1986                 kvm_iodevice_destructor(pos);
1987         }
1988         kfree(bus);
1989 }
1990
1991 /* kvm_io_bus_write - called under kvm->slots_lock */
1992 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
1993                      int len, const void *val)
1994 {
1995         int i;
1996         struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
1997         for (i = 0; i < bus->dev_count; i++)
1998                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1999                         return 0;
2000         return -EOPNOTSUPP;
2001 }
2002
2003 /* kvm_io_bus_read - called under kvm->slots_lock */
2004 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2005                     int len, void *val)
2006 {
2007         int i;
2008         struct kvm_io_bus *bus = rcu_dereference(kvm->buses[bus_idx]);
2009
2010         for (i = 0; i < bus->dev_count; i++)
2011                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2012                         return 0;
2013         return -EOPNOTSUPP;
2014 }
2015
2016 /* Caller must hold slots_lock. */
2017 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2018                             struct kvm_io_device *dev)
2019 {
2020         struct kvm_io_bus *new_bus, *bus;
2021
2022         bus = kvm->buses[bus_idx];
2023         if (bus->dev_count > NR_IOBUS_DEVS-1)
2024                 return -ENOSPC;
2025
2026         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2027         if (!new_bus)
2028                 return -ENOMEM;
2029         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2030         new_bus->devs[new_bus->dev_count++] = dev;
2031         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2032         synchronize_srcu_expedited(&kvm->srcu);
2033         kfree(bus);
2034
2035         return 0;
2036 }
2037
2038 /* Caller must hold slots_lock. */
2039 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2040                               struct kvm_io_device *dev)
2041 {
2042         int i, r;
2043         struct kvm_io_bus *new_bus, *bus;
2044
2045         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2046         if (!new_bus)
2047                 return -ENOMEM;
2048
2049         bus = kvm->buses[bus_idx];
2050         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2051
2052         r = -ENOENT;
2053         for (i = 0; i < new_bus->dev_count; i++)
2054                 if (new_bus->devs[i] == dev) {
2055                         r = 0;
2056                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2057                         break;
2058                 }
2059
2060         if (r) {
2061                 kfree(new_bus);
2062                 return r;
2063         }
2064
2065         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2066         synchronize_srcu_expedited(&kvm->srcu);
2067         kfree(bus);
2068         return r;
2069 }
2070
2071 static struct notifier_block kvm_cpu_notifier = {
2072         .notifier_call = kvm_cpu_hotplug,
2073         .priority = 20, /* must be > scheduler priority */
2074 };
2075
2076 static int vm_stat_get(void *_offset, u64 *val)
2077 {
2078         unsigned offset = (long)_offset;
2079         struct kvm *kvm;
2080
2081         *val = 0;
2082         spin_lock(&kvm_lock);
2083         list_for_each_entry(kvm, &vm_list, vm_list)
2084                 *val += *(u32 *)((void *)kvm + offset);
2085         spin_unlock(&kvm_lock);
2086         return 0;
2087 }
2088
2089 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2090
2091 static int vcpu_stat_get(void *_offset, u64 *val)
2092 {
2093         unsigned offset = (long)_offset;
2094         struct kvm *kvm;
2095         struct kvm_vcpu *vcpu;
2096         int i;
2097
2098         *val = 0;
2099         spin_lock(&kvm_lock);
2100         list_for_each_entry(kvm, &vm_list, vm_list)
2101                 kvm_for_each_vcpu(i, vcpu, kvm)
2102                         *val += *(u32 *)((void *)vcpu + offset);
2103
2104         spin_unlock(&kvm_lock);
2105         return 0;
2106 }
2107
2108 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2109
2110 static const struct file_operations *stat_fops[] = {
2111         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2112         [KVM_STAT_VM]   = &vm_stat_fops,
2113 };
2114
2115 static void kvm_init_debug(void)
2116 {
2117         struct kvm_stats_debugfs_item *p;
2118
2119         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2120         for (p = debugfs_entries; p->name; ++p)
2121                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2122                                                 (void *)(long)p->offset,
2123                                                 stat_fops[p->kind]);
2124 }
2125
2126 static void kvm_exit_debug(void)
2127 {
2128         struct kvm_stats_debugfs_item *p;
2129
2130         for (p = debugfs_entries; p->name; ++p)
2131                 debugfs_remove(p->dentry);
2132         debugfs_remove(kvm_debugfs_dir);
2133 }
2134
2135 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2136 {
2137         if (kvm_usage_count)
2138                 hardware_disable(NULL);
2139         return 0;
2140 }
2141
2142 static int kvm_resume(struct sys_device *dev)
2143 {
2144         if (kvm_usage_count)
2145                 hardware_enable(NULL);
2146         return 0;
2147 }
2148
2149 static struct sysdev_class kvm_sysdev_class = {
2150         .name = "kvm",
2151         .suspend = kvm_suspend,
2152         .resume = kvm_resume,
2153 };
2154
2155 static struct sys_device kvm_sysdev = {
2156         .id = 0,
2157         .cls = &kvm_sysdev_class,
2158 };
2159
2160 struct page *bad_page;
2161 pfn_t bad_pfn;
2162
2163 static inline
2164 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2165 {
2166         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2167 }
2168
2169 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2170 {
2171         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2172
2173         kvm_arch_vcpu_load(vcpu, cpu);
2174 }
2175
2176 static void kvm_sched_out(struct preempt_notifier *pn,
2177                           struct task_struct *next)
2178 {
2179         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2180
2181         kvm_arch_vcpu_put(vcpu);
2182 }
2183
2184 int kvm_init(void *opaque, unsigned int vcpu_size,
2185                   struct module *module)
2186 {
2187         int r;
2188         int cpu;
2189
2190         r = kvm_arch_init(opaque);
2191         if (r)
2192                 goto out_fail;
2193
2194         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2195
2196         if (bad_page == NULL) {
2197                 r = -ENOMEM;
2198                 goto out;
2199         }
2200
2201         bad_pfn = page_to_pfn(bad_page);
2202
2203         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2204                 r = -ENOMEM;
2205                 goto out_free_0;
2206         }
2207
2208         r = kvm_arch_hardware_setup();
2209         if (r < 0)
2210                 goto out_free_0a;
2211
2212         for_each_online_cpu(cpu) {
2213                 smp_call_function_single(cpu,
2214                                 kvm_arch_check_processor_compat,
2215                                 &r, 1);
2216                 if (r < 0)
2217                         goto out_free_1;
2218         }
2219
2220         r = register_cpu_notifier(&kvm_cpu_notifier);
2221         if (r)
2222                 goto out_free_2;
2223         register_reboot_notifier(&kvm_reboot_notifier);
2224
2225         r = sysdev_class_register(&kvm_sysdev_class);
2226         if (r)
2227                 goto out_free_3;
2228
2229         r = sysdev_register(&kvm_sysdev);
2230         if (r)
2231                 goto out_free_4;
2232
2233         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2234         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2235                                            __alignof__(struct kvm_vcpu),
2236                                            0, NULL);
2237         if (!kvm_vcpu_cache) {
2238                 r = -ENOMEM;
2239                 goto out_free_5;
2240         }
2241
2242         kvm_chardev_ops.owner = module;
2243         kvm_vm_fops.owner = module;
2244         kvm_vcpu_fops.owner = module;
2245
2246         r = misc_register(&kvm_dev);
2247         if (r) {
2248                 printk(KERN_ERR "kvm: misc device register failed\n");
2249                 goto out_free;
2250         }
2251
2252         kvm_preempt_ops.sched_in = kvm_sched_in;
2253         kvm_preempt_ops.sched_out = kvm_sched_out;
2254
2255         kvm_init_debug();
2256
2257         return 0;
2258
2259 out_free:
2260         kmem_cache_destroy(kvm_vcpu_cache);
2261 out_free_5:
2262         sysdev_unregister(&kvm_sysdev);
2263 out_free_4:
2264         sysdev_class_unregister(&kvm_sysdev_class);
2265 out_free_3:
2266         unregister_reboot_notifier(&kvm_reboot_notifier);
2267         unregister_cpu_notifier(&kvm_cpu_notifier);
2268 out_free_2:
2269 out_free_1:
2270         kvm_arch_hardware_unsetup();
2271 out_free_0a:
2272         free_cpumask_var(cpus_hardware_enabled);
2273 out_free_0:
2274         __free_page(bad_page);
2275 out:
2276         kvm_arch_exit();
2277 out_fail:
2278         return r;
2279 }
2280 EXPORT_SYMBOL_GPL(kvm_init);
2281
2282 void kvm_exit(void)
2283 {
2284         kvm_exit_debug();
2285         misc_deregister(&kvm_dev);
2286         kmem_cache_destroy(kvm_vcpu_cache);
2287         sysdev_unregister(&kvm_sysdev);
2288         sysdev_class_unregister(&kvm_sysdev_class);
2289         unregister_reboot_notifier(&kvm_reboot_notifier);
2290         unregister_cpu_notifier(&kvm_cpu_notifier);
2291         on_each_cpu(hardware_disable, NULL, 1);
2292         kvm_arch_hardware_unsetup();
2293         kvm_arch_exit();
2294         free_cpumask_var(cpus_hardware_enabled);
2295         __free_page(bad_page);
2296 }
2297 EXPORT_SYMBOL_GPL(kvm_exit);