virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/sysdev.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50
  51 #include <asm/processor.h>
  52 #include <asm/io.h>
  53 #include <asm/uaccess.h>
  54 #include <asm/pgtable.h>
  55 #include <asm-generic/bitops/le.h>
  56
  57 #include "coalesced_mmio.h"
  58 #include "async_pf.h"
  59
  60 #define CREATE_TRACE_POINTS
  61 #include <trace/events/kvm.h>
  62
  63 MODULE_AUTHOR("Qumranet");
  64 MODULE_LICENSE("GPL");
  65
  66 /*
  67  * Ordering of locks:
  68  *
  69  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  70  */
  71
  72 DEFINE_SPINLOCK(kvm_lock);
  73 LIST_HEAD(vm_list);
  74
  75 static cpumask_var_t cpus_hardware_enabled;
  76 static int kvm_usage_count = 0;
  77 static atomic_t hardware_enable_failed;
  78
  79 struct kmem_cache *kvm_vcpu_cache;
  80 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  81
  82 static __read_mostly struct preempt_ops kvm_preempt_ops;
  83
  84 struct dentry *kvm_debugfs_dir;
  85
  86 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  87                            unsigned long arg);
  88 static int hardware_enable_all(void);
  89 static void hardware_disable_all(void);
  90
  91 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  92
  93 static bool kvm_rebooting;
  94
  95 static bool largepages_enabled = true;
  96
  97 static struct page *hwpoison_page;
  98 static pfn_t hwpoison_pfn;
  99
 100 static struct page *fault_page;
 101 static pfn_t fault_pfn;
 102
 103 inline int kvm_is_mmio_pfn(pfn_t pfn)
 104 {
 105         if (pfn_valid(pfn)) {
 106                 struct page *page = compound_head(pfn_to_page(pfn));
 107                 return PageReserved(page);
 108         }
 109
 110         return true;
 111 }
 112
 113 /*
 114  * Switches to specified vcpu, until a matching vcpu_put()
 115  */
 116 void vcpu_load(struct kvm_vcpu *vcpu)
 117 {
 118         int cpu;
 119
 120         mutex_lock(&vcpu->mutex);
 121         cpu = get_cpu();
 122         preempt_notifier_register(&vcpu->preempt_notifier);
 123         kvm_arch_vcpu_load(vcpu, cpu);
 124         put_cpu();
 125 }
 126
 127 void vcpu_put(struct kvm_vcpu *vcpu)
 128 {
 129         preempt_disable();
 130         kvm_arch_vcpu_put(vcpu);
 131         preempt_notifier_unregister(&vcpu->preempt_notifier);
 132         preempt_enable();
 133         mutex_unlock(&vcpu->mutex);
 134 }
 135
 136 static void ack_flush(void *_completed)
 137 {
 138 }
 139
 140 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 141 {
 142         int i, cpu, me;
 143         cpumask_var_t cpus;
 144         bool called = true;
 145         struct kvm_vcpu *vcpu;
 146
 147         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 148
 149         raw_spin_lock(&kvm->requests_lock);
 150         me = smp_processor_id();
 151         kvm_for_each_vcpu(i, vcpu, kvm) {
 152                 if (kvm_make_check_request(req, vcpu))
 153                         continue;
 154                 cpu = vcpu->cpu;
 155                 if (cpus != NULL && cpu != -1 && cpu != me)
 156                         cpumask_set_cpu(cpu, cpus);
 157         }
 158         if (unlikely(cpus == NULL))
 159                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 160         else if (!cpumask_empty(cpus))
 161                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 162         else
 163                 called = false;
 164         raw_spin_unlock(&kvm->requests_lock);
 165         free_cpumask_var(cpus);
 166         return called;
 167 }
 168
 169 void kvm_flush_remote_tlbs(struct kvm *kvm)
 170 {
 171         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 172                 ++kvm->stat.remote_tlb_flush;
 173 }
 174
 175 void kvm_reload_remote_mmus(struct kvm *kvm)
 176 {
 177         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 178 }
 179
 180 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 181 {
 182         struct page *page;
 183         int r;
 184
 185         mutex_init(&vcpu->mutex);
 186         vcpu->cpu = -1;
 187         vcpu->kvm = kvm;
 188         vcpu->vcpu_id = id;
 189         init_waitqueue_head(&vcpu->wq);
 190         kvm_async_pf_vcpu_init(vcpu);
 191
 192         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 193         if (!page) {
 194                 r = -ENOMEM;
 195                 goto fail;
 196         }
 197         vcpu->run = page_address(page);
 198
 199         r = kvm_arch_vcpu_init(vcpu);
 200         if (r < 0)
 201                 goto fail_free_run;
 202         return 0;
 203
 204 fail_free_run:
 205         free_page((unsigned long)vcpu->run);
 206 fail:
 207         return r;
 208 }
 209 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 210
 211 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 212 {
 213         kvm_arch_vcpu_uninit(vcpu);
 214         free_page((unsigned long)vcpu->run);
 215 }
 216 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 217
 218 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 219 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 220 {
 221         return container_of(mn, struct kvm, mmu_notifier);
 222 }
 223
 224 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 225                                              struct mm_struct *mm,
 226                                              unsigned long address)
 227 {
 228         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 229         int need_tlb_flush, idx;
 230
 231         /*
 232          * When ->invalidate_page runs, the linux pte has been zapped
 233          * already but the page is still allocated until
 234          * ->invalidate_page returns. So if we increase the sequence
 235          * here the kvm page fault will notice if the spte can't be
 236          * established because the page is going to be freed. If
 237          * instead the kvm page fault establishes the spte before
 238          * ->invalidate_page runs, kvm_unmap_hva will release it
 239          * before returning.
 240          *
 241          * The sequence increase only need to be seen at spin_unlock
 242          * time, and not at spin_lock time.
 243          *
 244          * Increasing the sequence after the spin_unlock would be
 245          * unsafe because the kvm page fault could then establish the
 246          * pte after kvm_unmap_hva returned, without noticing the page
 247          * is going to be freed.
 248          */
 249         idx = srcu_read_lock(&kvm->srcu);
 250         spin_lock(&kvm->mmu_lock);
 251         kvm->mmu_notifier_seq++;
 252         need_tlb_flush = kvm_unmap_hva(kvm, address);
 253         spin_unlock(&kvm->mmu_lock);
 254         srcu_read_unlock(&kvm->srcu, idx);
 255
 256         /* we've to flush the tlb before the pages can be freed */
 257         if (need_tlb_flush)
 258                 kvm_flush_remote_tlbs(kvm);
 259
 260 }
 261
 262 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 263                                         struct mm_struct *mm,
 264                                         unsigned long address,
 265                                         pte_t pte)
 266 {
 267         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 268         int idx;
 269
 270         idx = srcu_read_lock(&kvm->srcu);
 271         spin_lock(&kvm->mmu_lock);
 272         kvm->mmu_notifier_seq++;
 273         kvm_set_spte_hva(kvm, address, pte);
 274         spin_unlock(&kvm->mmu_lock);
 275         srcu_read_unlock(&kvm->srcu, idx);
 276 }
 277
 278 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 279                                                     struct mm_struct *mm,
 280                                                     unsigned long start,
 281                                                     unsigned long end)
 282 {
 283         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 284         int need_tlb_flush = 0, idx;
 285
 286         idx = srcu_read_lock(&kvm->srcu);
 287         spin_lock(&kvm->mmu_lock);
 288         /*
 289          * The count increase must become visible at unlock time as no
 290          * spte can be established without taking the mmu_lock and
 291          * count is also read inside the mmu_lock critical section.
 292          */
 293         kvm->mmu_notifier_count++;
 294         for (; start < end; start += PAGE_SIZE)
 295                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 296         spin_unlock(&kvm->mmu_lock);
 297         srcu_read_unlock(&kvm->srcu, idx);
 298
 299         /* we've to flush the tlb before the pages can be freed */
 300         if (need_tlb_flush)
 301                 kvm_flush_remote_tlbs(kvm);
 302 }
 303
 304 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 305                                                   struct mm_struct *mm,
 306                                                   unsigned long start,
 307                                                   unsigned long end)
 308 {
 309         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 310
 311         spin_lock(&kvm->mmu_lock);
 312         /*
 313          * This sequence increase will notify the kvm page fault that
 314          * the page that is going to be mapped in the spte could have
 315          * been freed.
 316          */
 317         kvm->mmu_notifier_seq++;
 318         /*
 319          * The above sequence increase must be visible before the
 320          * below count decrease but both values are read by the kvm
 321          * page fault under mmu_lock spinlock so we don't need to add
 322          * a smb_wmb() here in between the two.
 323          */
 324         kvm->mmu_notifier_count--;
 325         spin_unlock(&kvm->mmu_lock);
 326
 327         BUG_ON(kvm->mmu_notifier_count < 0);
 328 }
 329
 330 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 331                                               struct mm_struct *mm,
 332                                               unsigned long address)
 333 {
 334         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 335         int young, idx;
 336
 337         idx = srcu_read_lock(&kvm->srcu);
 338         spin_lock(&kvm->mmu_lock);
 339         young = kvm_age_hva(kvm, address);
 340         spin_unlock(&kvm->mmu_lock);
 341         srcu_read_unlock(&kvm->srcu, idx);
 342
 343         if (young)
 344                 kvm_flush_remote_tlbs(kvm);
 345
 346         return young;
 347 }
 348
 349 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 350                                      struct mm_struct *mm)
 351 {
 352         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 353         int idx;
 354
 355         idx = srcu_read_lock(&kvm->srcu);
 356         kvm_arch_flush_shadow(kvm);
 357         srcu_read_unlock(&kvm->srcu, idx);
 358 }
 359
 360 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 361         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 362         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 363         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 364         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 365         .change_pte             = kvm_mmu_notifier_change_pte,
 366         .release                = kvm_mmu_notifier_release,
 367 };
 368
 369 static int kvm_init_mmu_notifier(struct kvm *kvm)
 370 {
 371         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 372         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 373 }
 374
 375 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 376
 377 static int kvm_init_mmu_notifier(struct kvm *kvm)
 378 {
 379         return 0;
 380 }
 381
 382 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 383
 384 static struct kvm *kvm_create_vm(void)
 385 {
 386         int r = 0, i;
 387         struct kvm *kvm = kvm_arch_create_vm();
 388
 389         if (IS_ERR(kvm))
 390                 goto out;
 391
 392         r = hardware_enable_all();
 393         if (r)
 394                 goto out_err_nodisable;
 395
 396 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 397         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 398         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 399 #endif
 400
 401         r = -ENOMEM;
 402         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 403         if (!kvm->memslots)
 404                 goto out_err;
 405         if (init_srcu_struct(&kvm->srcu))
 406                 goto out_err;
 407         for (i = 0; i < KVM_NR_BUSES; i++) {
 408                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 409                                         GFP_KERNEL);
 410                 if (!kvm->buses[i]) {
 411                         cleanup_srcu_struct(&kvm->srcu);
 412                         goto out_err;
 413                 }
 414         }
 415
 416         r = kvm_init_mmu_notifier(kvm);
 417         if (r) {
 418                 cleanup_srcu_struct(&kvm->srcu);
 419                 goto out_err;
 420         }
 421
 422         kvm->mm = current->mm;
 423         atomic_inc(&kvm->mm->mm_count);
 424         spin_lock_init(&kvm->mmu_lock);
 425         raw_spin_lock_init(&kvm->requests_lock);
 426         kvm_eventfd_init(kvm);
 427         mutex_init(&kvm->lock);
 428         mutex_init(&kvm->irq_lock);
 429         mutex_init(&kvm->slots_lock);
 430         atomic_set(&kvm->users_count, 1);
 431         spin_lock(&kvm_lock);
 432         list_add(&kvm->vm_list, &vm_list);
 433         spin_unlock(&kvm_lock);
 434 out:
 435         return kvm;
 436
 437 out_err:
 438         hardware_disable_all();
 439 out_err_nodisable:
 440         for (i = 0; i < KVM_NR_BUSES; i++)
 441                 kfree(kvm->buses[i]);
 442         kfree(kvm->memslots);
 443         kfree(kvm);
 444         return ERR_PTR(r);
 445 }
 446
 447 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 448 {
 449         if (!memslot->dirty_bitmap)
 450                 return;
 451
 452         vfree(memslot->dirty_bitmap);
 453         memslot->dirty_bitmap = NULL;
 454 }
 455
 456 /*
 457  * Free any memory in @free but not in @dont.
 458  */
 459 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 460                                   struct kvm_memory_slot *dont)
 461 {
 462         int i;
 463
 464         if (!dont || free->rmap != dont->rmap)
 465                 vfree(free->rmap);
 466
 467         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 468                 kvm_destroy_dirty_bitmap(free);
 469
 470
 471         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 472                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 473                         vfree(free->lpage_info[i]);
 474                         free->lpage_info[i] = NULL;
 475                 }
 476         }
 477
 478         free->npages = 0;
 479         free->rmap = NULL;
 480 }
 481
 482 void kvm_free_physmem(struct kvm *kvm)
 483 {
 484         int i;
 485         struct kvm_memslots *slots = kvm->memslots;
 486
 487         for (i = 0; i < slots->nmemslots; ++i)
 488                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 489
 490         kfree(kvm->memslots);
 491 }
 492
 493 static void kvm_destroy_vm(struct kvm *kvm)
 494 {
 495         int i;
 496         struct mm_struct *mm = kvm->mm;
 497
 498         kvm_arch_sync_events(kvm);
 499         spin_lock(&kvm_lock);
 500         list_del(&kvm->vm_list);
 501         spin_unlock(&kvm_lock);
 502         kvm_free_irq_routing(kvm);
 503         for (i = 0; i < KVM_NR_BUSES; i++)
 504                 kvm_io_bus_destroy(kvm->buses[i]);
 505         kvm_coalesced_mmio_free(kvm);
 506 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 507         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 508 #else
 509         kvm_arch_flush_shadow(kvm);
 510 #endif
 511         kvm_arch_destroy_vm(kvm);
 512         hardware_disable_all();
 513         mmdrop(mm);
 514 }
 515
 516 void kvm_get_kvm(struct kvm *kvm)
 517 {
 518         atomic_inc(&kvm->users_count);
 519 }
 520 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 521
 522 void kvm_put_kvm(struct kvm *kvm)
 523 {
 524         if (atomic_dec_and_test(&kvm->users_count))
 525                 kvm_destroy_vm(kvm);
 526 }
 527 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 528
 529
 530 static int kvm_vm_release(struct inode *inode, struct file *filp)
 531 {
 532         struct kvm *kvm = filp->private_data;
 533
 534         kvm_irqfd_release(kvm);
 535
 536         kvm_put_kvm(kvm);
 537         return 0;
 538 }
 539
 540 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 541 {
 542         unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
 543
 544         memslot->dirty_bitmap = vmalloc(dirty_bytes);
 545         if (!memslot->dirty_bitmap)
 546                 return -ENOMEM;
 547
 548         memset(memslot->dirty_bitmap, 0, dirty_bytes);
 549         return 0;
 550 }
 551
 552 /*
 553  * Allocate some memory and give it an address in the guest physical address
 554  * space.
 555  *
 556  * Discontiguous memory is allowed, mostly for framebuffers.
 557  *
 558  * Must be called holding mmap_sem for write.
 559  */
 560 int __kvm_set_memory_region(struct kvm *kvm,
 561                             struct kvm_userspace_memory_region *mem,
 562                             int user_alloc)
 563 {
 564         int r, flush_shadow = 0;
 565         gfn_t base_gfn;
 566         unsigned long npages;
 567         unsigned long i;
 568         struct kvm_memory_slot *memslot;
 569         struct kvm_memory_slot old, new;
 570         struct kvm_memslots *slots, *old_memslots;
 571
 572         r = -EINVAL;
 573         /* General sanity checks */
 574         if (mem->memory_size & (PAGE_SIZE - 1))
 575                 goto out;
 576         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 577                 goto out;
 578         if (user_alloc && (mem->userspace_addr & (PAGE_SIZE - 1)))
 579                 goto out;
 580         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 581                 goto out;
 582         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 583                 goto out;
 584
 585         memslot = &kvm->memslots->memslots[mem->slot];
 586         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 587         npages = mem->memory_size >> PAGE_SHIFT;
 588
 589         r = -EINVAL;
 590         if (npages > KVM_MEM_MAX_NR_PAGES)
 591                 goto out;
 592
 593         if (!npages)
 594                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 595
 596         new = old = *memslot;
 597
 598         new.id = mem->slot;
 599         new.base_gfn = base_gfn;
 600         new.npages = npages;
 601         new.flags = mem->flags;
 602
 603         /* Disallow changing a memory slot's size. */
 604         r = -EINVAL;
 605         if (npages && old.npages && npages != old.npages)
 606                 goto out_free;
 607
 608         /* Check for overlaps */
 609         r = -EEXIST;
 610         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 611                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 612
 613                 if (s == memslot || !s->npages)
 614                         continue;
 615                 if (!((base_gfn + npages <= s->base_gfn) ||
 616                       (base_gfn >= s->base_gfn + s->npages)))
 617                         goto out_free;
 618         }
 619
 620         /* Free page dirty bitmap if unneeded */
 621         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 622                 new.dirty_bitmap = NULL;
 623
 624         r = -ENOMEM;
 625
 626         /* Allocate if a slot is being created */
 627 #ifndef CONFIG_S390
 628         if (npages && !new.rmap) {
 629                 new.rmap = vmalloc(npages * sizeof(*new.rmap));
 630
 631                 if (!new.rmap)
 632                         goto out_free;
 633
 634                 memset(new.rmap, 0, npages * sizeof(*new.rmap));
 635
 636                 new.user_alloc = user_alloc;
 637                 new.userspace_addr = mem->userspace_addr;
 638         }
 639         if (!npages)
 640                 goto skip_lpage;
 641
 642         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 643                 unsigned long ugfn;
 644                 unsigned long j;
 645                 int lpages;
 646                 int level = i + 2;
 647
 648                 /* Avoid unused variable warning if no large pages */
 649                 (void)level;
 650
 651                 if (new.lpage_info[i])
 652                         continue;
 653
 654                 lpages = 1 + ((base_gfn + npages - 1)
 655                              >> KVM_HPAGE_GFN_SHIFT(level));
 656                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 657
 658                 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
 659
 660                 if (!new.lpage_info[i])
 661                         goto out_free;
 662
 663                 memset(new.lpage_info[i], 0,
 664                        lpages * sizeof(*new.lpage_info[i]));
 665
 666                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 667                         new.lpage_info[i][0].write_count = 1;
 668                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 669                         new.lpage_info[i][lpages - 1].write_count = 1;
 670                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 671                 /*
 672                  * If the gfn and userspace address are not aligned wrt each
 673                  * other, or if explicitly asked to, disable large page
 674                  * support for this slot
 675                  */
 676                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 677                     !largepages_enabled)
 678                         for (j = 0; j < lpages; ++j)
 679                                 new.lpage_info[i][j].write_count = 1;
 680         }
 681
 682 skip_lpage:
 683
 684         /* Allocate page dirty bitmap if needed */
 685         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 686                 if (kvm_create_dirty_bitmap(&new) < 0)
 687                         goto out_free;
 688                 /* destroy any largepage mappings for dirty tracking */
 689                 if (old.npages)
 690                         flush_shadow = 1;
 691         }
 692 #else  /* not defined CONFIG_S390 */
 693         new.user_alloc = user_alloc;
 694         if (user_alloc)
 695                 new.userspace_addr = mem->userspace_addr;
 696 #endif /* not defined CONFIG_S390 */
 697
 698         if (!npages) {
 699                 r = -ENOMEM;
 700                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 701                 if (!slots)
 702                         goto out_free;
 703                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 704                 if (mem->slot >= slots->nmemslots)
 705                         slots->nmemslots = mem->slot + 1;
 706                 slots->generation++;
 707                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 708
 709                 old_memslots = kvm->memslots;
 710                 rcu_assign_pointer(kvm->memslots, slots);
 711                 synchronize_srcu_expedited(&kvm->srcu);
 712                 /* From this point no new shadow pages pointing to a deleted
 713                  * memslot will be created.
 714                  *
 715                  * validation of sp->gfn happens in:
 716                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 717                  *      - kvm_is_visible_gfn (mmu_check_roots)
 718                  */
 719                 kvm_arch_flush_shadow(kvm);
 720                 kfree(old_memslots);
 721         }
 722
 723         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 724         if (r)
 725                 goto out_free;
 726
 727         /* map the pages in iommu page table */
 728         if (npages) {
 729                 r = kvm_iommu_map_pages(kvm, &new);
 730                 if (r)
 731                         goto out_free;
 732         }
 733
 734         r = -ENOMEM;
 735         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 736         if (!slots)
 737                 goto out_free;
 738         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 739         if (mem->slot >= slots->nmemslots)
 740                 slots->nmemslots = mem->slot + 1;
 741         slots->generation++;
 742
 743         /* actual memory is freed via old in kvm_free_physmem_slot below */
 744         if (!npages) {
 745                 new.rmap = NULL;
 746                 new.dirty_bitmap = NULL;
 747                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 748                         new.lpage_info[i] = NULL;
 749         }
 750
 751         slots->memslots[mem->slot] = new;
 752         old_memslots = kvm->memslots;
 753         rcu_assign_pointer(kvm->memslots, slots);
 754         synchronize_srcu_expedited(&kvm->srcu);
 755
 756         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 757
 758         kvm_free_physmem_slot(&old, &new);
 759         kfree(old_memslots);
 760
 761         if (flush_shadow)
 762                 kvm_arch_flush_shadow(kvm);
 763
 764         return 0;
 765
 766 out_free:
 767         kvm_free_physmem_slot(&new, &old);
 768 out:
 769         return r;
 770
 771 }
 772 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 773
 774 int kvm_set_memory_region(struct kvm *kvm,
 775                           struct kvm_userspace_memory_region *mem,
 776                           int user_alloc)
 777 {
 778         int r;
 779
 780         mutex_lock(&kvm->slots_lock);
 781         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 782         mutex_unlock(&kvm->slots_lock);
 783         return r;
 784 }
 785 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 786
 787 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 788                                    struct
 789                                    kvm_userspace_memory_region *mem,
 790                                    int user_alloc)
 791 {
 792         if (mem->slot >= KVM_MEMORY_SLOTS)
 793                 return -EINVAL;
 794         return kvm_set_memory_region(kvm, mem, user_alloc);
 795 }
 796
 797 int kvm_get_dirty_log(struct kvm *kvm,
 798                         struct kvm_dirty_log *log, int *is_dirty)
 799 {
 800         struct kvm_memory_slot *memslot;
 801         int r, i;
 802         unsigned long n;
 803         unsigned long any = 0;
 804
 805         r = -EINVAL;
 806         if (log->slot >= KVM_MEMORY_SLOTS)
 807                 goto out;
 808
 809         memslot = &kvm->memslots->memslots[log->slot];
 810         r = -ENOENT;
 811         if (!memslot->dirty_bitmap)
 812                 goto out;
 813
 814         n = kvm_dirty_bitmap_bytes(memslot);
 815
 816         for (i = 0; !any && i < n/sizeof(long); ++i)
 817                 any = memslot->dirty_bitmap[i];
 818
 819         r = -EFAULT;
 820         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 821                 goto out;
 822
 823         if (any)
 824                 *is_dirty = 1;
 825
 826         r = 0;
 827 out:
 828         return r;
 829 }
 830
 831 void kvm_disable_largepages(void)
 832 {
 833         largepages_enabled = false;
 834 }
 835 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 836
 837 int is_error_page(struct page *page)
 838 {
 839         return page == bad_page || page == hwpoison_page || page == fault_page;
 840 }
 841 EXPORT_SYMBOL_GPL(is_error_page);
 842
 843 int is_error_pfn(pfn_t pfn)
 844 {
 845         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 846 }
 847 EXPORT_SYMBOL_GPL(is_error_pfn);
 848
 849 int is_hwpoison_pfn(pfn_t pfn)
 850 {
 851         return pfn == hwpoison_pfn;
 852 }
 853 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 854
 855 int is_fault_pfn(pfn_t pfn)
 856 {
 857         return pfn == fault_pfn;
 858 }
 859 EXPORT_SYMBOL_GPL(is_fault_pfn);
 860
 861 static inline unsigned long bad_hva(void)
 862 {
 863         return PAGE_OFFSET;
 864 }
 865
 866 int kvm_is_error_hva(unsigned long addr)
 867 {
 868         return addr == bad_hva();
 869 }
 870 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 871
 872 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 873                                                 gfn_t gfn)
 874 {
 875         int i;
 876
 877         for (i = 0; i < slots->nmemslots; ++i) {
 878                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 879
 880                 if (gfn >= memslot->base_gfn
 881                     && gfn < memslot->base_gfn + memslot->npages)
 882                         return memslot;
 883         }
 884         return NULL;
 885 }
 886
 887 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 888 {
 889         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 890 }
 891 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 892
 893 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 894 {
 895         int i;
 896         struct kvm_memslots *slots = kvm_memslots(kvm);
 897
 898         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 899                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 900
 901                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 902                         continue;
 903
 904                 if (gfn >= memslot->base_gfn
 905                     && gfn < memslot->base_gfn + memslot->npages)
 906                         return 1;
 907         }
 908         return 0;
 909 }
 910 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 911
 912 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 913 {
 914         struct vm_area_struct *vma;
 915         unsigned long addr, size;
 916
 917         size = PAGE_SIZE;
 918
 919         addr = gfn_to_hva(kvm, gfn);
 920         if (kvm_is_error_hva(addr))
 921                 return PAGE_SIZE;
 922
 923         down_read(&current->mm->mmap_sem);
 924         vma = find_vma(current->mm, addr);
 925         if (!vma)
 926                 goto out;
 927
 928         size = vma_kernel_pagesize(vma);
 929
 930 out:
 931         up_read(&current->mm->mmap_sem);
 932
 933         return size;
 934 }
 935
 936 int memslot_id(struct kvm *kvm, gfn_t gfn)
 937 {
 938         int i;
 939         struct kvm_memslots *slots = kvm_memslots(kvm);
 940         struct kvm_memory_slot *memslot = NULL;
 941
 942         for (i = 0; i < slots->nmemslots; ++i) {
 943                 memslot = &slots->memslots[i];
 944
 945                 if (gfn >= memslot->base_gfn
 946                     && gfn < memslot->base_gfn + memslot->npages)
 947                         break;
 948         }
 949
 950         return memslot - slots->memslots;
 951 }
 952
 953 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
 954                                      gfn_t *nr_pages)
 955 {
 956         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 957                 return bad_hva();
 958
 959         if (nr_pages)
 960                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
 961
 962         return gfn_to_hva_memslot(slot, gfn);
 963 }
 964
 965 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 966 {
 967         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
 968 }
 969 EXPORT_SYMBOL_GPL(gfn_to_hva);
 970
 971 static pfn_t get_fault_pfn(void)
 972 {
 973         get_page(fault_page);
 974         return fault_pfn;
 975 }
 976
 977 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
 978                         bool *async, bool write_fault, bool *writable)
 979 {
 980         struct page *page[1];
 981         int npages = 0;
 982         pfn_t pfn;
 983
 984         /* we can do it either atomically or asynchronously, not both */
 985         BUG_ON(atomic && async);
 986
 987         BUG_ON(!write_fault && !writable);
 988
 989         if (writable)
 990                 *writable = true;
 991
 992         if (atomic || async)
 993                 npages = __get_user_pages_fast(addr, 1, 1, page);
 994
 995         if (unlikely(npages != 1) && !atomic) {
 996                 might_sleep();
 997
 998                 if (writable)
 999                         *writable = write_fault;
1000
1001                 npages = get_user_pages_fast(addr, 1, write_fault, page);
1002
1003                 /* map read fault as writable if possible */
1004                 if (unlikely(!write_fault) && npages == 1) {
1005                         struct page *wpage[1];
1006
1007                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1008                         if (npages == 1) {
1009                                 *writable = true;
1010                                 put_page(page[0]);
1011                                 page[0] = wpage[0];
1012                         }
1013                         npages = 1;
1014                 }
1015         }
1016
1017         if (unlikely(npages != 1)) {
1018                 struct vm_area_struct *vma;
1019
1020                 if (atomic)
1021                         return get_fault_pfn();
1022
1023                 down_read(&current->mm->mmap_sem);
1024                 if (is_hwpoison_address(addr)) {
1025                         up_read(&current->mm->mmap_sem);
1026                         get_page(hwpoison_page);
1027                         return page_to_pfn(hwpoison_page);
1028                 }
1029
1030                 vma = find_vma_intersection(current->mm, addr, addr+1);
1031
1032                 if (vma == NULL)
1033                         pfn = get_fault_pfn();
1034                 else if ((vma->vm_flags & VM_PFNMAP)) {
1035                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1036                                 vma->vm_pgoff;
1037                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1038                 } else {
1039                         if (async && (vma->vm_flags & VM_WRITE))
1040                                 *async = true;
1041                         pfn = get_fault_pfn();
1042                 }
1043                 up_read(&current->mm->mmap_sem);
1044         } else
1045                 pfn = page_to_pfn(page[0]);
1046
1047         return pfn;
1048 }
1049
1050 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1051 {
1052         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1053 }
1054 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1055
1056 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1057                           bool write_fault, bool *writable)
1058 {
1059         unsigned long addr;
1060
1061         if (async)
1062                 *async = false;
1063
1064         addr = gfn_to_hva(kvm, gfn);
1065         if (kvm_is_error_hva(addr)) {
1066                 get_page(bad_page);
1067                 return page_to_pfn(bad_page);
1068         }
1069
1070         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1071 }
1072
1073 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1074 {
1075         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1076 }
1077 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1078
1079 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1080                        bool write_fault, bool *writable)
1081 {
1082         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1083 }
1084 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1085
1086 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1087 {
1088         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1089 }
1090 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1091
1092 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1093                       bool *writable)
1094 {
1095         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1096 }
1097 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1098
1099 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1100                          struct kvm_memory_slot *slot, gfn_t gfn)
1101 {
1102         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1103         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1104 }
1105
1106 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1107                                                                   int nr_pages)
1108 {
1109         unsigned long addr;
1110         gfn_t entry;
1111
1112         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1113         if (kvm_is_error_hva(addr))
1114                 return -1;
1115
1116         if (entry < nr_pages)
1117                 return 0;
1118
1119         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1120 }
1121 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1122
1123 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1124 {
1125         pfn_t pfn;
1126
1127         pfn = gfn_to_pfn(kvm, gfn);
1128         if (!kvm_is_mmio_pfn(pfn))
1129                 return pfn_to_page(pfn);
1130
1131         WARN_ON(kvm_is_mmio_pfn(pfn));
1132
1133         get_page(bad_page);
1134         return bad_page;
1135 }
1136
1137 EXPORT_SYMBOL_GPL(gfn_to_page);
1138
1139 void kvm_release_page_clean(struct page *page)
1140 {
1141         kvm_release_pfn_clean(page_to_pfn(page));
1142 }
1143 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1144
1145 void kvm_release_pfn_clean(pfn_t pfn)
1146 {
1147         if (!kvm_is_mmio_pfn(pfn))
1148                 put_page(pfn_to_page(pfn));
1149 }
1150 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1151
1152 void kvm_release_page_dirty(struct page *page)
1153 {
1154         kvm_release_pfn_dirty(page_to_pfn(page));
1155 }
1156 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1157
1158 void kvm_release_pfn_dirty(pfn_t pfn)
1159 {
1160         kvm_set_pfn_dirty(pfn);
1161         kvm_release_pfn_clean(pfn);
1162 }
1163 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1164
1165 void kvm_set_page_dirty(struct page *page)
1166 {
1167         kvm_set_pfn_dirty(page_to_pfn(page));
1168 }
1169 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1170
1171 void kvm_set_pfn_dirty(pfn_t pfn)
1172 {
1173         if (!kvm_is_mmio_pfn(pfn)) {
1174                 struct page *page = pfn_to_page(pfn);
1175                 if (!PageReserved(page))
1176                         SetPageDirty(page);
1177         }
1178 }
1179 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1180
1181 void kvm_set_pfn_accessed(pfn_t pfn)
1182 {
1183         if (!kvm_is_mmio_pfn(pfn))
1184                 mark_page_accessed(pfn_to_page(pfn));
1185 }
1186 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1187
1188 void kvm_get_pfn(pfn_t pfn)
1189 {
1190         if (!kvm_is_mmio_pfn(pfn))
1191                 get_page(pfn_to_page(pfn));
1192 }
1193 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1194
1195 static int next_segment(unsigned long len, int offset)
1196 {
1197         if (len > PAGE_SIZE - offset)
1198                 return PAGE_SIZE - offset;
1199         else
1200                 return len;
1201 }
1202
1203 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1204                         int len)
1205 {
1206         int r;
1207         unsigned long addr;
1208
1209         addr = gfn_to_hva(kvm, gfn);
1210         if (kvm_is_error_hva(addr))
1211                 return -EFAULT;
1212         r = copy_from_user(data, (void __user *)addr + offset, len);
1213         if (r)
1214                 return -EFAULT;
1215         return 0;
1216 }
1217 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1218
1219 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1220 {
1221         gfn_t gfn = gpa >> PAGE_SHIFT;
1222         int seg;
1223         int offset = offset_in_page(gpa);
1224         int ret;
1225
1226         while ((seg = next_segment(len, offset)) != 0) {
1227                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1228                 if (ret < 0)
1229                         return ret;
1230                 offset = 0;
1231                 len -= seg;
1232                 data += seg;
1233                 ++gfn;
1234         }
1235         return 0;
1236 }
1237 EXPORT_SYMBOL_GPL(kvm_read_guest);
1238
1239 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1240                           unsigned long len)
1241 {
1242         int r;
1243         unsigned long addr;
1244         gfn_t gfn = gpa >> PAGE_SHIFT;
1245         int offset = offset_in_page(gpa);
1246
1247         addr = gfn_to_hva(kvm, gfn);
1248         if (kvm_is_error_hva(addr))
1249                 return -EFAULT;
1250         pagefault_disable();
1251         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1252         pagefault_enable();
1253         if (r)
1254                 return -EFAULT;
1255         return 0;
1256 }
1257 EXPORT_SYMBOL(kvm_read_guest_atomic);
1258
1259 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1260                          int offset, int len)
1261 {
1262         int r;
1263         unsigned long addr;
1264
1265         addr = gfn_to_hva(kvm, gfn);
1266         if (kvm_is_error_hva(addr))
1267                 return -EFAULT;
1268         r = copy_to_user((void __user *)addr + offset, data, len);
1269         if (r)
1270                 return -EFAULT;
1271         mark_page_dirty(kvm, gfn);
1272         return 0;
1273 }
1274 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1275
1276 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1277                     unsigned long len)
1278 {
1279         gfn_t gfn = gpa >> PAGE_SHIFT;
1280         int seg;
1281         int offset = offset_in_page(gpa);
1282         int ret;
1283
1284         while ((seg = next_segment(len, offset)) != 0) {
1285                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1286                 if (ret < 0)
1287                         return ret;
1288                 offset = 0;
1289                 len -= seg;
1290                 data += seg;
1291                 ++gfn;
1292         }
1293         return 0;
1294 }
1295
1296 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1297                               gpa_t gpa)
1298 {
1299         struct kvm_memslots *slots = kvm_memslots(kvm);
1300         int offset = offset_in_page(gpa);
1301         gfn_t gfn = gpa >> PAGE_SHIFT;
1302
1303         ghc->gpa = gpa;
1304         ghc->generation = slots->generation;
1305         ghc->memslot = __gfn_to_memslot(slots, gfn);
1306         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1307         if (!kvm_is_error_hva(ghc->hva))
1308                 ghc->hva += offset;
1309         else
1310                 return -EFAULT;
1311
1312         return 0;
1313 }
1314 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1315
1316 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1317                            void *data, unsigned long len)
1318 {
1319         struct kvm_memslots *slots = kvm_memslots(kvm);
1320         int r;
1321
1322         if (slots->generation != ghc->generation)
1323                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1324
1325         if (kvm_is_error_hva(ghc->hva))
1326                 return -EFAULT;
1327
1328         r = copy_to_user((void __user *)ghc->hva, data, len);
1329         if (r)
1330                 return -EFAULT;
1331         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1332
1333         return 0;
1334 }
1335 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1336
1337 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1338 {
1339         return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1340 }
1341 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1342
1343 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1344 {
1345         gfn_t gfn = gpa >> PAGE_SHIFT;
1346         int seg;
1347         int offset = offset_in_page(gpa);
1348         int ret;
1349
1350         while ((seg = next_segment(len, offset)) != 0) {
1351                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1352                 if (ret < 0)
1353                         return ret;
1354                 offset = 0;
1355                 len -= seg;
1356                 ++gfn;
1357         }
1358         return 0;
1359 }
1360 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1361
1362 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1363                              gfn_t gfn)
1364 {
1365         if (memslot && memslot->dirty_bitmap) {
1366                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1367
1368                 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1369         }
1370 }
1371
1372 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1373 {
1374         struct kvm_memory_slot *memslot;
1375
1376         memslot = gfn_to_memslot(kvm, gfn);
1377         mark_page_dirty_in_slot(kvm, memslot, gfn);
1378 }
1379
1380 /*
1381  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1382  */
1383 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1384 {
1385         DEFINE_WAIT(wait);
1386
1387         for (;;) {
1388                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1389
1390                 if (kvm_arch_vcpu_runnable(vcpu)) {
1391                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1392                         break;
1393                 }
1394                 if (kvm_cpu_has_pending_timer(vcpu))
1395                         break;
1396                 if (signal_pending(current))
1397                         break;
1398
1399                 schedule();
1400         }
1401
1402         finish_wait(&vcpu->wq, &wait);
1403 }
1404
1405 void kvm_resched(struct kvm_vcpu *vcpu)
1406 {
1407         if (!need_resched())
1408                 return;
1409         cond_resched();
1410 }
1411 EXPORT_SYMBOL_GPL(kvm_resched);
1412
1413 void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1414 {
1415         ktime_t expires;
1416         DEFINE_WAIT(wait);
1417
1418         prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1419
1420         /* Sleep for 100 us, and hope lock-holder got scheduled */
1421         expires = ktime_add_ns(ktime_get(), 100000UL);
1422         schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1423
1424         finish_wait(&vcpu->wq, &wait);
1425 }
1426 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1427
1428 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1429 {
1430         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1431         struct page *page;
1432
1433         if (vmf->pgoff == 0)
1434                 page = virt_to_page(vcpu->run);
1435 #ifdef CONFIG_X86
1436         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1437                 page = virt_to_page(vcpu->arch.pio_data);
1438 #endif
1439 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1440         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1441                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1442 #endif
1443         else
1444                 return VM_FAULT_SIGBUS;
1445         get_page(page);
1446         vmf->page = page;
1447         return 0;
1448 }
1449
1450 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1451         .fault = kvm_vcpu_fault,
1452 };
1453
1454 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1455 {
1456         vma->vm_ops = &kvm_vcpu_vm_ops;
1457         return 0;
1458 }
1459
1460 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1461 {
1462         struct kvm_vcpu *vcpu = filp->private_data;
1463
1464         kvm_put_kvm(vcpu->kvm);
1465         return 0;
1466 }
1467
1468 static struct file_operations kvm_vcpu_fops = {
1469         .release        = kvm_vcpu_release,
1470         .unlocked_ioctl = kvm_vcpu_ioctl,
1471         .compat_ioctl   = kvm_vcpu_ioctl,
1472         .mmap           = kvm_vcpu_mmap,
1473         .llseek         = noop_llseek,
1474 };
1475
1476 /*
1477  * Allocates an inode for the vcpu.
1478  */
1479 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1480 {
1481         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1482 }
1483
1484 /*
1485  * Creates some virtual cpus.  Good luck creating more than one.
1486  */
1487 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1488 {
1489         int r;
1490         struct kvm_vcpu *vcpu, *v;
1491
1492         vcpu = kvm_arch_vcpu_create(kvm, id);
1493         if (IS_ERR(vcpu))
1494                 return PTR_ERR(vcpu);
1495
1496         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1497
1498         r = kvm_arch_vcpu_setup(vcpu);
1499         if (r)
1500                 return r;
1501
1502         mutex_lock(&kvm->lock);
1503         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1504                 r = -EINVAL;
1505                 goto vcpu_destroy;
1506         }
1507
1508         kvm_for_each_vcpu(r, v, kvm)
1509                 if (v->vcpu_id == id) {
1510                         r = -EEXIST;
1511                         goto vcpu_destroy;
1512                 }
1513
1514         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1515
1516         /* Now it's all set up, let userspace reach it */
1517         kvm_get_kvm(kvm);
1518         r = create_vcpu_fd(vcpu);
1519         if (r < 0) {
1520                 kvm_put_kvm(kvm);
1521                 goto vcpu_destroy;
1522         }
1523
1524         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1525         smp_wmb();
1526         atomic_inc(&kvm->online_vcpus);
1527
1528 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1529         if (kvm->bsp_vcpu_id == id)
1530                 kvm->bsp_vcpu = vcpu;
1531 #endif
1532         mutex_unlock(&kvm->lock);
1533         return r;
1534
1535 vcpu_destroy:
1536         mutex_unlock(&kvm->lock);
1537         kvm_arch_vcpu_destroy(vcpu);
1538         return r;
1539 }
1540
1541 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1542 {
1543         if (sigset) {
1544                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1545                 vcpu->sigset_active = 1;
1546                 vcpu->sigset = *sigset;
1547         } else
1548                 vcpu->sigset_active = 0;
1549         return 0;
1550 }
1551
1552 static long kvm_vcpu_ioctl(struct file *filp,
1553                            unsigned int ioctl, unsigned long arg)
1554 {
1555         struct kvm_vcpu *vcpu = filp->private_data;
1556         void __user *argp = (void __user *)arg;
1557         int r;
1558         struct kvm_fpu *fpu = NULL;
1559         struct kvm_sregs *kvm_sregs = NULL;
1560
1561         if (vcpu->kvm->mm != current->mm)
1562                 return -EIO;
1563
1564 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1565         /*
1566          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1567          * so vcpu_load() would break it.
1568          */
1569         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1570                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1571 #endif
1572
1573
1574         vcpu_load(vcpu);
1575         switch (ioctl) {
1576         case KVM_RUN:
1577                 r = -EINVAL;
1578                 if (arg)
1579                         goto out;
1580                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1581                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1582                 break;
1583         case KVM_GET_REGS: {
1584                 struct kvm_regs *kvm_regs;
1585
1586                 r = -ENOMEM;
1587                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1588                 if (!kvm_regs)
1589                         goto out;
1590                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1591                 if (r)
1592                         goto out_free1;
1593                 r = -EFAULT;
1594                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1595                         goto out_free1;
1596                 r = 0;
1597 out_free1:
1598                 kfree(kvm_regs);
1599                 break;
1600         }
1601         case KVM_SET_REGS: {
1602                 struct kvm_regs *kvm_regs;
1603
1604                 r = -ENOMEM;
1605                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1606                 if (!kvm_regs)
1607                         goto out;
1608                 r = -EFAULT;
1609                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1610                         goto out_free2;
1611                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1612                 if (r)
1613                         goto out_free2;
1614                 r = 0;
1615 out_free2:
1616                 kfree(kvm_regs);
1617                 break;
1618         }
1619         case KVM_GET_SREGS: {
1620                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1621                 r = -ENOMEM;
1622                 if (!kvm_sregs)
1623                         goto out;
1624                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1625                 if (r)
1626                         goto out;
1627                 r = -EFAULT;
1628                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1629                         goto out;
1630                 r = 0;
1631                 break;
1632         }
1633         case KVM_SET_SREGS: {
1634                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1635                 r = -ENOMEM;
1636                 if (!kvm_sregs)
1637                         goto out;
1638                 r = -EFAULT;
1639                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1640                         goto out;
1641                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1642                 if (r)
1643                         goto out;
1644                 r = 0;
1645                 break;
1646         }
1647         case KVM_GET_MP_STATE: {
1648                 struct kvm_mp_state mp_state;
1649
1650                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1651                 if (r)
1652                         goto out;
1653                 r = -EFAULT;
1654                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1655                         goto out;
1656                 r = 0;
1657                 break;
1658         }
1659         case KVM_SET_MP_STATE: {
1660                 struct kvm_mp_state mp_state;
1661
1662                 r = -EFAULT;
1663                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1664                         goto out;
1665                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1666                 if (r)
1667                         goto out;
1668                 r = 0;
1669                 break;
1670         }
1671         case KVM_TRANSLATE: {
1672                 struct kvm_translation tr;
1673
1674                 r = -EFAULT;
1675                 if (copy_from_user(&tr, argp, sizeof tr))
1676                         goto out;
1677                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1678                 if (r)
1679                         goto out;
1680                 r = -EFAULT;
1681                 if (copy_to_user(argp, &tr, sizeof tr))
1682                         goto out;
1683                 r = 0;
1684                 break;
1685         }
1686         case KVM_SET_GUEST_DEBUG: {
1687                 struct kvm_guest_debug dbg;
1688
1689                 r = -EFAULT;
1690                 if (copy_from_user(&dbg, argp, sizeof dbg))
1691                         goto out;
1692                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1693                 if (r)
1694                         goto out;
1695                 r = 0;
1696                 break;
1697         }
1698         case KVM_SET_SIGNAL_MASK: {
1699                 struct kvm_signal_mask __user *sigmask_arg = argp;
1700                 struct kvm_signal_mask kvm_sigmask;
1701                 sigset_t sigset, *p;
1702
1703                 p = NULL;
1704                 if (argp) {
1705                         r = -EFAULT;
1706                         if (copy_from_user(&kvm_sigmask, argp,
1707                                            sizeof kvm_sigmask))
1708                                 goto out;
1709                         r = -EINVAL;
1710                         if (kvm_sigmask.len != sizeof sigset)
1711                                 goto out;
1712                         r = -EFAULT;
1713                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1714                                            sizeof sigset))
1715                                 goto out;
1716                         p = &sigset;
1717                 }
1718                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1719                 break;
1720         }
1721         case KVM_GET_FPU: {
1722                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1723                 r = -ENOMEM;
1724                 if (!fpu)
1725                         goto out;
1726                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1727                 if (r)
1728                         goto out;
1729                 r = -EFAULT;
1730                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1731                         goto out;
1732                 r = 0;
1733                 break;
1734         }
1735         case KVM_SET_FPU: {
1736                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1737                 r = -ENOMEM;
1738                 if (!fpu)
1739                         goto out;
1740                 r = -EFAULT;
1741                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1742                         goto out;
1743                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1744                 if (r)
1745                         goto out;
1746                 r = 0;
1747                 break;
1748         }
1749         default:
1750                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1751         }
1752 out:
1753         vcpu_put(vcpu);
1754         kfree(fpu);
1755         kfree(kvm_sregs);
1756         return r;
1757 }
1758
1759 static long kvm_vm_ioctl(struct file *filp,
1760                            unsigned int ioctl, unsigned long arg)
1761 {
1762         struct kvm *kvm = filp->private_data;
1763         void __user *argp = (void __user *)arg;
1764         int r;
1765
1766         if (kvm->mm != current->mm)
1767                 return -EIO;
1768         switch (ioctl) {
1769         case KVM_CREATE_VCPU:
1770                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1771                 if (r < 0)
1772                         goto out;
1773                 break;
1774         case KVM_SET_USER_MEMORY_REGION: {
1775                 struct kvm_userspace_memory_region kvm_userspace_mem;
1776
1777                 r = -EFAULT;
1778                 if (copy_from_user(&kvm_userspace_mem, argp,
1779                                                 sizeof kvm_userspace_mem))
1780                         goto out;
1781
1782                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1783                 if (r)
1784                         goto out;
1785                 break;
1786         }
1787         case KVM_GET_DIRTY_LOG: {
1788                 struct kvm_dirty_log log;
1789
1790                 r = -EFAULT;
1791                 if (copy_from_user(&log, argp, sizeof log))
1792                         goto out;
1793                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1794                 if (r)
1795                         goto out;
1796                 break;
1797         }
1798 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1799         case KVM_REGISTER_COALESCED_MMIO: {
1800                 struct kvm_coalesced_mmio_zone zone;
1801                 r = -EFAULT;
1802                 if (copy_from_user(&zone, argp, sizeof zone))
1803                         goto out;
1804                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1805                 if (r)
1806                         goto out;
1807                 r = 0;
1808                 break;
1809         }
1810         case KVM_UNREGISTER_COALESCED_MMIO: {
1811                 struct kvm_coalesced_mmio_zone zone;
1812                 r = -EFAULT;
1813                 if (copy_from_user(&zone, argp, sizeof zone))
1814                         goto out;
1815                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1816                 if (r)
1817                         goto out;
1818                 r = 0;
1819                 break;
1820         }
1821 #endif
1822         case KVM_IRQFD: {
1823                 struct kvm_irqfd data;
1824
1825                 r = -EFAULT;
1826                 if (copy_from_user(&data, argp, sizeof data))
1827                         goto out;
1828                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
1829                 break;
1830         }
1831         case KVM_IOEVENTFD: {
1832                 struct kvm_ioeventfd data;
1833
1834                 r = -EFAULT;
1835                 if (copy_from_user(&data, argp, sizeof data))
1836                         goto out;
1837                 r = kvm_ioeventfd(kvm, &data);
1838                 break;
1839         }
1840 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1841         case KVM_SET_BOOT_CPU_ID:
1842                 r = 0;
1843                 mutex_lock(&kvm->lock);
1844                 if (atomic_read(&kvm->online_vcpus) != 0)
1845                         r = -EBUSY;
1846                 else
1847                         kvm->bsp_vcpu_id = arg;
1848                 mutex_unlock(&kvm->lock);
1849                 break;
1850 #endif
1851         default:
1852                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1853                 if (r == -ENOTTY)
1854                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
1855         }
1856 out:
1857         return r;
1858 }
1859
1860 #ifdef CONFIG_COMPAT
1861 struct compat_kvm_dirty_log {
1862         __u32 slot;
1863         __u32 padding1;
1864         union {
1865                 compat_uptr_t dirty_bitmap; /* one bit per page */
1866                 __u64 padding2;
1867         };
1868 };
1869
1870 static long kvm_vm_compat_ioctl(struct file *filp,
1871                            unsigned int ioctl, unsigned long arg)
1872 {
1873         struct kvm *kvm = filp->private_data;
1874         int r;
1875
1876         if (kvm->mm != current->mm)
1877                 return -EIO;
1878         switch (ioctl) {
1879         case KVM_GET_DIRTY_LOG: {
1880                 struct compat_kvm_dirty_log compat_log;
1881                 struct kvm_dirty_log log;
1882
1883                 r = -EFAULT;
1884                 if (copy_from_user(&compat_log, (void __user *)arg,
1885                                    sizeof(compat_log)))
1886                         goto out;
1887                 log.slot         = compat_log.slot;
1888                 log.padding1     = compat_log.padding1;
1889                 log.padding2     = compat_log.padding2;
1890                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1891
1892                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1893                 if (r)
1894                         goto out;
1895                 break;
1896         }
1897         default:
1898                 r = kvm_vm_ioctl(filp, ioctl, arg);
1899         }
1900
1901 out:
1902         return r;
1903 }
1904 #endif
1905
1906 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1907 {
1908         struct page *page[1];
1909         unsigned long addr;
1910         int npages;
1911         gfn_t gfn = vmf->pgoff;
1912         struct kvm *kvm = vma->vm_file->private_data;
1913
1914         addr = gfn_to_hva(kvm, gfn);
1915         if (kvm_is_error_hva(addr))
1916                 return VM_FAULT_SIGBUS;
1917
1918         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1919                                 NULL);
1920         if (unlikely(npages != 1))
1921                 return VM_FAULT_SIGBUS;
1922
1923         vmf->page = page[0];
1924         return 0;
1925 }
1926
1927 static const struct vm_operations_struct kvm_vm_vm_ops = {
1928         .fault = kvm_vm_fault,
1929 };
1930
1931 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1932 {
1933         vma->vm_ops = &kvm_vm_vm_ops;
1934         return 0;
1935 }
1936
1937 static struct file_operations kvm_vm_fops = {
1938         .release        = kvm_vm_release,
1939         .unlocked_ioctl = kvm_vm_ioctl,
1940 #ifdef CONFIG_COMPAT
1941         .compat_ioctl   = kvm_vm_compat_ioctl,
1942 #endif
1943         .mmap           = kvm_vm_mmap,
1944         .llseek         = noop_llseek,
1945 };
1946
1947 static int kvm_dev_ioctl_create_vm(void)
1948 {
1949         int fd, r;
1950         struct kvm *kvm;
1951
1952         kvm = kvm_create_vm();
1953         if (IS_ERR(kvm))
1954                 return PTR_ERR(kvm);
1955 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1956         r = kvm_coalesced_mmio_init(kvm);
1957         if (r < 0) {
1958                 kvm_put_kvm(kvm);
1959                 return r;
1960         }
1961 #endif
1962         fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
1963         if (fd < 0)
1964                 kvm_put_kvm(kvm);
1965
1966         return fd;
1967 }
1968
1969 static long kvm_dev_ioctl_check_extension_generic(long arg)
1970 {
1971         switch (arg) {
1972         case KVM_CAP_USER_MEMORY:
1973         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
1974         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1975 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1976         case KVM_CAP_SET_BOOT_CPU_ID:
1977 #endif
1978         case KVM_CAP_INTERNAL_ERROR_DATA:
1979                 return 1;
1980 #ifdef CONFIG_HAVE_KVM_IRQCHIP
1981         case KVM_CAP_IRQ_ROUTING:
1982                 return KVM_MAX_IRQ_ROUTES;
1983 #endif
1984         default:
1985                 break;
1986         }
1987         return kvm_dev_ioctl_check_extension(arg);
1988 }
1989
1990 static long kvm_dev_ioctl(struct file *filp,
1991                           unsigned int ioctl, unsigned long arg)
1992 {
1993         long r = -EINVAL;
1994
1995         switch (ioctl) {
1996         case KVM_GET_API_VERSION:
1997                 r = -EINVAL;
1998                 if (arg)
1999                         goto out;
2000                 r = KVM_API_VERSION;
2001                 break;
2002         case KVM_CREATE_VM:
2003                 r = -EINVAL;
2004                 if (arg)
2005                         goto out;
2006                 r = kvm_dev_ioctl_create_vm();
2007                 break;
2008         case KVM_CHECK_EXTENSION:
2009                 r = kvm_dev_ioctl_check_extension_generic(arg);
2010                 break;
2011         case KVM_GET_VCPU_MMAP_SIZE:
2012                 r = -EINVAL;
2013                 if (arg)
2014                         goto out;
2015                 r = PAGE_SIZE;     /* struct kvm_run */
2016 #ifdef CONFIG_X86
2017                 r += PAGE_SIZE;    /* pio data page */
2018 #endif
2019 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2020                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2021 #endif
2022                 break;
2023         case KVM_TRACE_ENABLE:
2024         case KVM_TRACE_PAUSE:
2025         case KVM_TRACE_DISABLE:
2026                 r = -EOPNOTSUPP;
2027                 break;
2028         default:
2029                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2030         }
2031 out:
2032         return r;
2033 }
2034
2035 static struct file_operations kvm_chardev_ops = {
2036         .unlocked_ioctl = kvm_dev_ioctl,
2037         .compat_ioctl   = kvm_dev_ioctl,
2038         .llseek         = noop_llseek,
2039 };
2040
2041 static struct miscdevice kvm_dev = {
2042         KVM_MINOR,
2043         "kvm",
2044         &kvm_chardev_ops,
2045 };
2046
2047 static void hardware_enable(void *junk)
2048 {
2049         int cpu = raw_smp_processor_id();
2050         int r;
2051
2052         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2053                 return;
2054
2055         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2056
2057         r = kvm_arch_hardware_enable(NULL);
2058
2059         if (r) {
2060                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2061                 atomic_inc(&hardware_enable_failed);
2062                 printk(KERN_INFO "kvm: enabling virtualization on "
2063                                  "CPU%d failed\n", cpu);
2064         }
2065 }
2066
2067 static void hardware_disable(void *junk)
2068 {
2069         int cpu = raw_smp_processor_id();
2070
2071         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2072                 return;
2073         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2074         kvm_arch_hardware_disable(NULL);
2075 }
2076
2077 static void hardware_disable_all_nolock(void)
2078 {
2079         BUG_ON(!kvm_usage_count);
2080
2081         kvm_usage_count--;
2082         if (!kvm_usage_count)
2083                 on_each_cpu(hardware_disable, NULL, 1);
2084 }
2085
2086 static void hardware_disable_all(void)
2087 {
2088         spin_lock(&kvm_lock);
2089         hardware_disable_all_nolock();
2090         spin_unlock(&kvm_lock);
2091 }
2092
2093 static int hardware_enable_all(void)
2094 {
2095         int r = 0;
2096
2097         spin_lock(&kvm_lock);
2098
2099         kvm_usage_count++;
2100         if (kvm_usage_count == 1) {
2101                 atomic_set(&hardware_enable_failed, 0);
2102                 on_each_cpu(hardware_enable, NULL, 1);
2103
2104                 if (atomic_read(&hardware_enable_failed)) {
2105                         hardware_disable_all_nolock();
2106                         r = -EBUSY;
2107                 }
2108         }
2109
2110         spin_unlock(&kvm_lock);
2111
2112         return r;
2113 }
2114
2115 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2116                            void *v)
2117 {
2118         int cpu = (long)v;
2119
2120         if (!kvm_usage_count)
2121                 return NOTIFY_OK;
2122
2123         val &= ~CPU_TASKS_FROZEN;
2124         switch (val) {
2125         case CPU_DYING:
2126                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2127                        cpu);
2128                 hardware_disable(NULL);
2129                 break;
2130         case CPU_STARTING:
2131                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2132                        cpu);
2133                 spin_lock(&kvm_lock);
2134                 hardware_enable(NULL);
2135                 spin_unlock(&kvm_lock);
2136                 break;
2137         }
2138         return NOTIFY_OK;
2139 }
2140
2141
2142 asmlinkage void kvm_handle_fault_on_reboot(void)
2143 {
2144         if (kvm_rebooting) {
2145                 /* spin while reset goes on */
2146                 local_irq_enable();
2147                 while (true)
2148                         cpu_relax();
2149         }
2150         /* Fault while not rebooting.  We want the trace. */
2151         BUG();
2152 }
2153 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
2154
2155 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2156                       void *v)
2157 {
2158         /*
2159          * Some (well, at least mine) BIOSes hang on reboot if
2160          * in vmx root mode.
2161          *
2162          * And Intel TXT required VMX off for all cpu when system shutdown.
2163          */
2164         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2165         kvm_rebooting = true;
2166         on_each_cpu(hardware_disable, NULL, 1);
2167         return NOTIFY_OK;
2168 }
2169
2170 static struct notifier_block kvm_reboot_notifier = {
2171         .notifier_call = kvm_reboot,
2172         .priority = 0,
2173 };
2174
2175 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2176 {
2177         int i;
2178
2179         for (i = 0; i < bus->dev_count; i++) {
2180                 struct kvm_io_device *pos = bus->devs[i];
2181
2182                 kvm_iodevice_destructor(pos);
2183         }
2184         kfree(bus);
2185 }
2186
2187 /* kvm_io_bus_write - called under kvm->slots_lock */
2188 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2189                      int len, const void *val)
2190 {
2191         int i;
2192         struct kvm_io_bus *bus;
2193
2194         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2195         for (i = 0; i < bus->dev_count; i++)
2196                 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
2197                         return 0;
2198         return -EOPNOTSUPP;
2199 }
2200
2201 /* kvm_io_bus_read - called under kvm->slots_lock */
2202 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2203                     int len, void *val)
2204 {
2205         int i;
2206         struct kvm_io_bus *bus;
2207
2208         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2209         for (i = 0; i < bus->dev_count; i++)
2210                 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
2211                         return 0;
2212         return -EOPNOTSUPP;
2213 }
2214
2215 /* Caller must hold slots_lock. */
2216 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2217                             struct kvm_io_device *dev)
2218 {
2219         struct kvm_io_bus *new_bus, *bus;
2220
2221         bus = kvm->buses[bus_idx];
2222         if (bus->dev_count > NR_IOBUS_DEVS-1)
2223                 return -ENOSPC;
2224
2225         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2226         if (!new_bus)
2227                 return -ENOMEM;
2228         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2229         new_bus->devs[new_bus->dev_count++] = dev;
2230         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2231         synchronize_srcu_expedited(&kvm->srcu);
2232         kfree(bus);
2233
2234         return 0;
2235 }
2236
2237 /* Caller must hold slots_lock. */
2238 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2239                               struct kvm_io_device *dev)
2240 {
2241         int i, r;
2242         struct kvm_io_bus *new_bus, *bus;
2243
2244         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2245         if (!new_bus)
2246                 return -ENOMEM;
2247
2248         bus = kvm->buses[bus_idx];
2249         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2250
2251         r = -ENOENT;
2252         for (i = 0; i < new_bus->dev_count; i++)
2253                 if (new_bus->devs[i] == dev) {
2254                         r = 0;
2255                         new_bus->devs[i] = new_bus->devs[--new_bus->dev_count];
2256                         break;
2257                 }
2258
2259         if (r) {
2260                 kfree(new_bus);
2261                 return r;
2262         }
2263
2264         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2265         synchronize_srcu_expedited(&kvm->srcu);
2266         kfree(bus);
2267         return r;
2268 }
2269
2270 static struct notifier_block kvm_cpu_notifier = {
2271         .notifier_call = kvm_cpu_hotplug,
2272 };
2273
2274 static int vm_stat_get(void *_offset, u64 *val)
2275 {
2276         unsigned offset = (long)_offset;
2277         struct kvm *kvm;
2278
2279         *val = 0;
2280         spin_lock(&kvm_lock);
2281         list_for_each_entry(kvm, &vm_list, vm_list)
2282                 *val += *(u32 *)((void *)kvm + offset);
2283         spin_unlock(&kvm_lock);
2284         return 0;
2285 }
2286
2287 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2288
2289 static int vcpu_stat_get(void *_offset, u64 *val)
2290 {
2291         unsigned offset = (long)_offset;
2292         struct kvm *kvm;
2293         struct kvm_vcpu *vcpu;
2294         int i;
2295
2296         *val = 0;
2297         spin_lock(&kvm_lock);
2298         list_for_each_entry(kvm, &vm_list, vm_list)
2299                 kvm_for_each_vcpu(i, vcpu, kvm)
2300                         *val += *(u32 *)((void *)vcpu + offset);
2301
2302         spin_unlock(&kvm_lock);
2303         return 0;
2304 }
2305
2306 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2307
2308 static const struct file_operations *stat_fops[] = {
2309         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2310         [KVM_STAT_VM]   = &vm_stat_fops,
2311 };
2312
2313 static void kvm_init_debug(void)
2314 {
2315         struct kvm_stats_debugfs_item *p;
2316
2317         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2318         for (p = debugfs_entries; p->name; ++p)
2319                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2320                                                 (void *)(long)p->offset,
2321                                                 stat_fops[p->kind]);
2322 }
2323
2324 static void kvm_exit_debug(void)
2325 {
2326         struct kvm_stats_debugfs_item *p;
2327
2328         for (p = debugfs_entries; p->name; ++p)
2329                 debugfs_remove(p->dentry);
2330         debugfs_remove(kvm_debugfs_dir);
2331 }
2332
2333 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2334 {
2335         if (kvm_usage_count)
2336                 hardware_disable(NULL);
2337         return 0;
2338 }
2339
2340 static int kvm_resume(struct sys_device *dev)
2341 {
2342         if (kvm_usage_count) {
2343                 WARN_ON(spin_is_locked(&kvm_lock));
2344                 hardware_enable(NULL);
2345         }
2346         return 0;
2347 }
2348
2349 static struct sysdev_class kvm_sysdev_class = {
2350         .name = "kvm",
2351         .suspend = kvm_suspend,
2352         .resume = kvm_resume,
2353 };
2354
2355 static struct sys_device kvm_sysdev = {
2356         .id = 0,
2357         .cls = &kvm_sysdev_class,
2358 };
2359
2360 struct page *bad_page;
2361 pfn_t bad_pfn;
2362
2363 static inline
2364 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2365 {
2366         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2367 }
2368
2369 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2370 {
2371         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2372
2373         kvm_arch_vcpu_load(vcpu, cpu);
2374 }
2375
2376 static void kvm_sched_out(struct preempt_notifier *pn,
2377                           struct task_struct *next)
2378 {
2379         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2380
2381         kvm_arch_vcpu_put(vcpu);
2382 }
2383
2384 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2385                   struct module *module)
2386 {
2387         int r;
2388         int cpu;
2389
2390         r = kvm_arch_init(opaque);
2391         if (r)
2392                 goto out_fail;
2393
2394         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2395
2396         if (bad_page == NULL) {
2397                 r = -ENOMEM;
2398                 goto out;
2399         }
2400
2401         bad_pfn = page_to_pfn(bad_page);
2402
2403         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2404
2405         if (hwpoison_page == NULL) {
2406                 r = -ENOMEM;
2407                 goto out_free_0;
2408         }
2409
2410         hwpoison_pfn = page_to_pfn(hwpoison_page);
2411
2412         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2413
2414         if (fault_page == NULL) {
2415                 r = -ENOMEM;
2416                 goto out_free_0;
2417         }
2418
2419         fault_pfn = page_to_pfn(fault_page);
2420
2421         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2422                 r = -ENOMEM;
2423                 goto out_free_0;
2424         }
2425
2426         r = kvm_arch_hardware_setup();
2427         if (r < 0)
2428                 goto out_free_0a;
2429
2430         for_each_online_cpu(cpu) {
2431                 smp_call_function_single(cpu,
2432                                 kvm_arch_check_processor_compat,
2433                                 &r, 1);
2434                 if (r < 0)
2435                         goto out_free_1;
2436         }
2437
2438         r = register_cpu_notifier(&kvm_cpu_notifier);
2439         if (r)
2440                 goto out_free_2;
2441         register_reboot_notifier(&kvm_reboot_notifier);
2442
2443         r = sysdev_class_register(&kvm_sysdev_class);
2444         if (r)
2445                 goto out_free_3;
2446
2447         r = sysdev_register(&kvm_sysdev);
2448         if (r)
2449                 goto out_free_4;
2450
2451         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2452         if (!vcpu_align)
2453                 vcpu_align = __alignof__(struct kvm_vcpu);
2454         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2455                                            0, NULL);
2456         if (!kvm_vcpu_cache) {
2457                 r = -ENOMEM;
2458                 goto out_free_5;
2459         }
2460
2461         r = kvm_async_pf_init();
2462         if (r)
2463                 goto out_free;
2464
2465         kvm_chardev_ops.owner = module;
2466         kvm_vm_fops.owner = module;
2467         kvm_vcpu_fops.owner = module;
2468
2469         r = misc_register(&kvm_dev);
2470         if (r) {
2471                 printk(KERN_ERR "kvm: misc device register failed\n");
2472                 goto out_unreg;
2473         }
2474
2475         kvm_preempt_ops.sched_in = kvm_sched_in;
2476         kvm_preempt_ops.sched_out = kvm_sched_out;
2477
2478         kvm_init_debug();
2479
2480         return 0;
2481
2482 out_unreg:
2483         kvm_async_pf_deinit();
2484 out_free:
2485         kmem_cache_destroy(kvm_vcpu_cache);
2486 out_free_5:
2487         sysdev_unregister(&kvm_sysdev);
2488 out_free_4:
2489         sysdev_class_unregister(&kvm_sysdev_class);
2490 out_free_3:
2491         unregister_reboot_notifier(&kvm_reboot_notifier);
2492         unregister_cpu_notifier(&kvm_cpu_notifier);
2493 out_free_2:
2494 out_free_1:
2495         kvm_arch_hardware_unsetup();
2496 out_free_0a:
2497         free_cpumask_var(cpus_hardware_enabled);
2498 out_free_0:
2499         if (fault_page)
2500                 __free_page(fault_page);
2501         if (hwpoison_page)
2502                 __free_page(hwpoison_page);
2503         __free_page(bad_page);
2504 out:
2505         kvm_arch_exit();
2506 out_fail:
2507         return r;
2508 }
2509 EXPORT_SYMBOL_GPL(kvm_init);
2510
2511 void kvm_exit(void)
2512 {
2513         kvm_exit_debug();
2514         misc_deregister(&kvm_dev);
2515         kmem_cache_destroy(kvm_vcpu_cache);
2516         kvm_async_pf_deinit();
2517         sysdev_unregister(&kvm_sysdev);
2518         sysdev_class_unregister(&kvm_sysdev_class);
2519         unregister_reboot_notifier(&kvm_reboot_notifier);
2520         unregister_cpu_notifier(&kvm_cpu_notifier);
2521         on_each_cpu(hardware_disable, NULL, 1);
2522         kvm_arch_hardware_unsetup();
2523         kvm_arch_exit();
2524         free_cpumask_var(cpus_hardware_enabled);
2525         __free_page(hwpoison_page);
2526         __free_page(bad_page);
2527 }
2528 EXPORT_SYMBOL_GPL(kvm_exit);