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/syscore_ops.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 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_RAW_SPINLOCK(kvm_lock);
  74 LIST_HEAD(vm_list);
  75
  76 static cpumask_var_t cpus_hardware_enabled;
  77 static int kvm_usage_count = 0;
  78 static atomic_t hardware_enable_failed;
  79
  80 struct kmem_cache *kvm_vcpu_cache;
  81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  82
  83 static __read_mostly struct preempt_ops kvm_preempt_ops;
  84
  85 struct dentry *kvm_debugfs_dir;
  86
  87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  88                            unsigned long arg);
  89 #ifdef CONFIG_COMPAT
  90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  91                                   unsigned long arg);
  92 #endif
  93 static int hardware_enable_all(void);
  94 static void hardware_disable_all(void);
  95
  96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  97
  98 bool kvm_rebooting;
  99 EXPORT_SYMBOL_GPL(kvm_rebooting);
 100
 101 static bool largepages_enabled = true;
 102
 103 static struct page *hwpoison_page;
 104 static pfn_t hwpoison_pfn;
 105
 106 struct page *fault_page;
 107 pfn_t fault_pfn;
 108
 109 inline int kvm_is_mmio_pfn(pfn_t pfn)
 110 {
 111         if (pfn_valid(pfn)) {
 112                 int reserved;
 113                 struct page *tail = pfn_to_page(pfn);
 114                 struct page *head = compound_trans_head(tail);
 115                 reserved = PageReserved(head);
 116                 if (head != tail) {
 117                         /*
 118                          * "head" is not a dangling pointer
 119                          * (compound_trans_head takes care of that)
 120                          * but the hugepage may have been splitted
 121                          * from under us (and we may not hold a
 122                          * reference count on the head page so it can
 123                          * be reused before we run PageReferenced), so
 124                          * we've to check PageTail before returning
 125                          * what we just read.
 126                          */
 127                         smp_rmb();
 128                         if (PageTail(tail))
 129                                 return reserved;
 130                 }
 131                 return PageReserved(tail);
 132         }
 133
 134         return true;
 135 }
 136
 137 /*
 138  * Switches to specified vcpu, until a matching vcpu_put()
 139  */
 140 void vcpu_load(struct kvm_vcpu *vcpu)
 141 {
 142         int cpu;
 143
 144         mutex_lock(&vcpu->mutex);
 145         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 146                 /* The thread running this VCPU changed. */
 147                 struct pid *oldpid = vcpu->pid;
 148                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 149                 rcu_assign_pointer(vcpu->pid, newpid);
 150                 synchronize_rcu();
 151                 put_pid(oldpid);
 152         }
 153         cpu = get_cpu();
 154         preempt_notifier_register(&vcpu->preempt_notifier);
 155         kvm_arch_vcpu_load(vcpu, cpu);
 156         put_cpu();
 157 }
 158
 159 void vcpu_put(struct kvm_vcpu *vcpu)
 160 {
 161         preempt_disable();
 162         kvm_arch_vcpu_put(vcpu);
 163         preempt_notifier_unregister(&vcpu->preempt_notifier);
 164         preempt_enable();
 165         mutex_unlock(&vcpu->mutex);
 166 }
 167
 168 static void ack_flush(void *_completed)
 169 {
 170 }
 171
 172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 173 {
 174         int i, cpu, me;
 175         cpumask_var_t cpus;
 176         bool called = true;
 177         struct kvm_vcpu *vcpu;
 178
 179         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 180
 181         me = get_cpu();
 182         kvm_for_each_vcpu(i, vcpu, kvm) {
 183                 kvm_make_request(req, vcpu);
 184                 cpu = vcpu->cpu;
 185
 186                 /* Set ->requests bit before we read ->mode */
 187                 smp_mb();
 188
 189                 if (cpus != NULL && cpu != -1 && cpu != me &&
 190                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 191                         cpumask_set_cpu(cpu, cpus);
 192         }
 193         if (unlikely(cpus == NULL))
 194                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 195         else if (!cpumask_empty(cpus))
 196                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 197         else
 198                 called = false;
 199         put_cpu();
 200         free_cpumask_var(cpus);
 201         return called;
 202 }
 203
 204 void kvm_flush_remote_tlbs(struct kvm *kvm)
 205 {
 206         int dirty_count = kvm->tlbs_dirty;
 207
 208         smp_mb();
 209         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 210                 ++kvm->stat.remote_tlb_flush;
 211         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 212 }
 213
 214 void kvm_reload_remote_mmus(struct kvm *kvm)
 215 {
 216         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 217 }
 218
 219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 220 {
 221         struct page *page;
 222         int r;
 223
 224         mutex_init(&vcpu->mutex);
 225         vcpu->cpu = -1;
 226         vcpu->kvm = kvm;
 227         vcpu->vcpu_id = id;
 228         vcpu->pid = NULL;
 229         init_waitqueue_head(&vcpu->wq);
 230         kvm_async_pf_vcpu_init(vcpu);
 231
 232         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 233         if (!page) {
 234                 r = -ENOMEM;
 235                 goto fail;
 236         }
 237         vcpu->run = page_address(page);
 238
 239         r = kvm_arch_vcpu_init(vcpu);
 240         if (r < 0)
 241                 goto fail_free_run;
 242         return 0;
 243
 244 fail_free_run:
 245         free_page((unsigned long)vcpu->run);
 246 fail:
 247         return r;
 248 }
 249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 250
 251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 252 {
 253         put_pid(vcpu->pid);
 254         kvm_arch_vcpu_uninit(vcpu);
 255         free_page((unsigned long)vcpu->run);
 256 }
 257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 258
 259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 261 {
 262         return container_of(mn, struct kvm, mmu_notifier);
 263 }
 264
 265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 266                                              struct mm_struct *mm,
 267                                              unsigned long address)
 268 {
 269         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 270         int need_tlb_flush, idx;
 271
 272         /*
 273          * When ->invalidate_page runs, the linux pte has been zapped
 274          * already but the page is still allocated until
 275          * ->invalidate_page returns. So if we increase the sequence
 276          * here the kvm page fault will notice if the spte can't be
 277          * established because the page is going to be freed. If
 278          * instead the kvm page fault establishes the spte before
 279          * ->invalidate_page runs, kvm_unmap_hva will release it
 280          * before returning.
 281          *
 282          * The sequence increase only need to be seen at spin_unlock
 283          * time, and not at spin_lock time.
 284          *
 285          * Increasing the sequence after the spin_unlock would be
 286          * unsafe because the kvm page fault could then establish the
 287          * pte after kvm_unmap_hva returned, without noticing the page
 288          * is going to be freed.
 289          */
 290         idx = srcu_read_lock(&kvm->srcu);
 291         spin_lock(&kvm->mmu_lock);
 292
 293         kvm->mmu_notifier_seq++;
 294         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 295         /* we've to flush the tlb before the pages can be freed */
 296         if (need_tlb_flush)
 297                 kvm_flush_remote_tlbs(kvm);
 298
 299         spin_unlock(&kvm->mmu_lock);
 300         srcu_read_unlock(&kvm->srcu, idx);
 301 }
 302
 303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 304                                         struct mm_struct *mm,
 305                                         unsigned long address,
 306                                         pte_t pte)
 307 {
 308         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 309         int idx;
 310
 311         idx = srcu_read_lock(&kvm->srcu);
 312         spin_lock(&kvm->mmu_lock);
 313         kvm->mmu_notifier_seq++;
 314         kvm_set_spte_hva(kvm, address, pte);
 315         spin_unlock(&kvm->mmu_lock);
 316         srcu_read_unlock(&kvm->srcu, idx);
 317 }
 318
 319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 320                                                     struct mm_struct *mm,
 321                                                     unsigned long start,
 322                                                     unsigned long end)
 323 {
 324         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 325         int need_tlb_flush = 0, idx;
 326
 327         idx = srcu_read_lock(&kvm->srcu);
 328         spin_lock(&kvm->mmu_lock);
 329         /*
 330          * The count increase must become visible at unlock time as no
 331          * spte can be established without taking the mmu_lock and
 332          * count is also read inside the mmu_lock critical section.
 333          */
 334         kvm->mmu_notifier_count++;
 335         for (; start < end; start += PAGE_SIZE)
 336                 need_tlb_flush |= kvm_unmap_hva(kvm, start);
 337         need_tlb_flush |= kvm->tlbs_dirty;
 338         /* we've to flush the tlb before the pages can be freed */
 339         if (need_tlb_flush)
 340                 kvm_flush_remote_tlbs(kvm);
 341
 342         spin_unlock(&kvm->mmu_lock);
 343         srcu_read_unlock(&kvm->srcu, idx);
 344 }
 345
 346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 347                                                   struct mm_struct *mm,
 348                                                   unsigned long start,
 349                                                   unsigned long end)
 350 {
 351         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 352
 353         spin_lock(&kvm->mmu_lock);
 354         /*
 355          * This sequence increase will notify the kvm page fault that
 356          * the page that is going to be mapped in the spte could have
 357          * been freed.
 358          */
 359         kvm->mmu_notifier_seq++;
 360         /*
 361          * The above sequence increase must be visible before the
 362          * below count decrease but both values are read by the kvm
 363          * page fault under mmu_lock spinlock so we don't need to add
 364          * a smb_wmb() here in between the two.
 365          */
 366         kvm->mmu_notifier_count--;
 367         spin_unlock(&kvm->mmu_lock);
 368
 369         BUG_ON(kvm->mmu_notifier_count < 0);
 370 }
 371
 372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 373                                               struct mm_struct *mm,
 374                                               unsigned long address)
 375 {
 376         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 377         int young, idx;
 378
 379         idx = srcu_read_lock(&kvm->srcu);
 380         spin_lock(&kvm->mmu_lock);
 381
 382         young = kvm_age_hva(kvm, address);
 383         if (young)
 384                 kvm_flush_remote_tlbs(kvm);
 385
 386         spin_unlock(&kvm->mmu_lock);
 387         srcu_read_unlock(&kvm->srcu, idx);
 388
 389         return young;
 390 }
 391
 392 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 393                                        struct mm_struct *mm,
 394                                        unsigned long address)
 395 {
 396         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 397         int young, idx;
 398
 399         idx = srcu_read_lock(&kvm->srcu);
 400         spin_lock(&kvm->mmu_lock);
 401         young = kvm_test_age_hva(kvm, address);
 402         spin_unlock(&kvm->mmu_lock);
 403         srcu_read_unlock(&kvm->srcu, idx);
 404
 405         return young;
 406 }
 407
 408 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 409                                      struct mm_struct *mm)
 410 {
 411         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 412         int idx;
 413
 414         idx = srcu_read_lock(&kvm->srcu);
 415         kvm_arch_flush_shadow(kvm);
 416         srcu_read_unlock(&kvm->srcu, idx);
 417 }
 418
 419 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 420         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 421         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 422         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 423         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 424         .test_young             = kvm_mmu_notifier_test_young,
 425         .change_pte             = kvm_mmu_notifier_change_pte,
 426         .release                = kvm_mmu_notifier_release,
 427 };
 428
 429 static int kvm_init_mmu_notifier(struct kvm *kvm)
 430 {
 431         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 432         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 433 }
 434
 435 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 436
 437 static int kvm_init_mmu_notifier(struct kvm *kvm)
 438 {
 439         return 0;
 440 }
 441
 442 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 443
 444 static struct kvm *kvm_create_vm(void)
 445 {
 446         int r, i;
 447         struct kvm *kvm = kvm_arch_alloc_vm();
 448
 449         if (!kvm)
 450                 return ERR_PTR(-ENOMEM);
 451
 452         r = kvm_arch_init_vm(kvm);
 453         if (r)
 454                 goto out_err_nodisable;
 455
 456         r = hardware_enable_all();
 457         if (r)
 458                 goto out_err_nodisable;
 459
 460 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 461         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 462         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 463 #endif
 464
 465         r = -ENOMEM;
 466         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 467         if (!kvm->memslots)
 468                 goto out_err_nosrcu;
 469         if (init_srcu_struct(&kvm->srcu))
 470                 goto out_err_nosrcu;
 471         for (i = 0; i < KVM_NR_BUSES; i++) {
 472                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 473                                         GFP_KERNEL);
 474                 if (!kvm->buses[i])
 475                         goto out_err;
 476         }
 477
 478         spin_lock_init(&kvm->mmu_lock);
 479         kvm->mm = current->mm;
 480         atomic_inc(&kvm->mm->mm_count);
 481         kvm_eventfd_init(kvm);
 482         mutex_init(&kvm->lock);
 483         mutex_init(&kvm->irq_lock);
 484         mutex_init(&kvm->slots_lock);
 485         atomic_set(&kvm->users_count, 1);
 486
 487         r = kvm_init_mmu_notifier(kvm);
 488         if (r)
 489                 goto out_err;
 490
 491         raw_spin_lock(&kvm_lock);
 492         list_add(&kvm->vm_list, &vm_list);
 493         raw_spin_unlock(&kvm_lock);
 494
 495         return kvm;
 496
 497 out_err:
 498         cleanup_srcu_struct(&kvm->srcu);
 499 out_err_nosrcu:
 500         hardware_disable_all();
 501 out_err_nodisable:
 502         for (i = 0; i < KVM_NR_BUSES; i++)
 503                 kfree(kvm->buses[i]);
 504         kfree(kvm->memslots);
 505         kvm_arch_free_vm(kvm);
 506         return ERR_PTR(r);
 507 }
 508
 509 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 510 {
 511         if (!memslot->dirty_bitmap)
 512                 return;
 513
 514         if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
 515                 vfree(memslot->dirty_bitmap_head);
 516         else
 517                 kfree(memslot->dirty_bitmap_head);
 518
 519         memslot->dirty_bitmap = NULL;
 520         memslot->dirty_bitmap_head = NULL;
 521 }
 522
 523 /*
 524  * Free any memory in @free but not in @dont.
 525  */
 526 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 527                                   struct kvm_memory_slot *dont)
 528 {
 529         int i;
 530
 531         if (!dont || free->rmap != dont->rmap)
 532                 vfree(free->rmap);
 533
 534         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 535                 kvm_destroy_dirty_bitmap(free);
 536
 537
 538         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 539                 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
 540                         vfree(free->lpage_info[i]);
 541                         free->lpage_info[i] = NULL;
 542                 }
 543         }
 544
 545         free->npages = 0;
 546         free->rmap = NULL;
 547 }
 548
 549 void kvm_free_physmem(struct kvm *kvm)
 550 {
 551         int i;
 552         struct kvm_memslots *slots = kvm->memslots;
 553
 554         for (i = 0; i < slots->nmemslots; ++i)
 555                 kvm_free_physmem_slot(&slots->memslots[i], NULL);
 556
 557         kfree(kvm->memslots);
 558 }
 559
 560 static void kvm_destroy_vm(struct kvm *kvm)
 561 {
 562         int i;
 563         struct mm_struct *mm = kvm->mm;
 564
 565         kvm_arch_sync_events(kvm);
 566         raw_spin_lock(&kvm_lock);
 567         list_del(&kvm->vm_list);
 568         raw_spin_unlock(&kvm_lock);
 569         kvm_free_irq_routing(kvm);
 570         for (i = 0; i < KVM_NR_BUSES; i++)
 571                 kvm_io_bus_destroy(kvm->buses[i]);
 572         kvm_coalesced_mmio_free(kvm);
 573 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 574         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 575 #else
 576         kvm_arch_flush_shadow(kvm);
 577 #endif
 578         kvm_arch_destroy_vm(kvm);
 579         kvm_free_physmem(kvm);
 580         cleanup_srcu_struct(&kvm->srcu);
 581         kvm_arch_free_vm(kvm);
 582         hardware_disable_all();
 583         mmdrop(mm);
 584 }
 585
 586 void kvm_get_kvm(struct kvm *kvm)
 587 {
 588         atomic_inc(&kvm->users_count);
 589 }
 590 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 591
 592 void kvm_put_kvm(struct kvm *kvm)
 593 {
 594         if (atomic_dec_and_test(&kvm->users_count))
 595                 kvm_destroy_vm(kvm);
 596 }
 597 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 598
 599
 600 static int kvm_vm_release(struct inode *inode, struct file *filp)
 601 {
 602         struct kvm *kvm = filp->private_data;
 603
 604         kvm_irqfd_release(kvm);
 605
 606         kvm_put_kvm(kvm);
 607         return 0;
 608 }
 609
 610 #ifndef CONFIG_S390
 611 /*
 612  * Allocation size is twice as large as the actual dirty bitmap size.
 613  * This makes it possible to do double buffering: see x86's
 614  * kvm_vm_ioctl_get_dirty_log().
 615  */
 616 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 617 {
 618         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 619
 620         if (dirty_bytes > PAGE_SIZE)
 621                 memslot->dirty_bitmap = vzalloc(dirty_bytes);
 622         else
 623                 memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
 624
 625         if (!memslot->dirty_bitmap)
 626                 return -ENOMEM;
 627
 628         memslot->dirty_bitmap_head = memslot->dirty_bitmap;
 629         return 0;
 630 }
 631 #endif /* !CONFIG_S390 */
 632
 633 /*
 634  * Allocate some memory and give it an address in the guest physical address
 635  * space.
 636  *
 637  * Discontiguous memory is allowed, mostly for framebuffers.
 638  *
 639  * Must be called holding mmap_sem for write.
 640  */
 641 int __kvm_set_memory_region(struct kvm *kvm,
 642                             struct kvm_userspace_memory_region *mem,
 643                             int user_alloc)
 644 {
 645         int r;
 646         gfn_t base_gfn;
 647         unsigned long npages;
 648         unsigned long i;
 649         struct kvm_memory_slot *memslot;
 650         struct kvm_memory_slot old, new;
 651         struct kvm_memslots *slots, *old_memslots;
 652
 653         r = -EINVAL;
 654         /* General sanity checks */
 655         if (mem->memory_size & (PAGE_SIZE - 1))
 656                 goto out;
 657         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 658                 goto out;
 659         /* We can read the guest memory with __xxx_user() later on. */
 660         if (user_alloc &&
 661             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 662              !access_ok(VERIFY_WRITE,
 663                         (void __user *)(unsigned long)mem->userspace_addr,
 664                         mem->memory_size)))
 665                 goto out;
 666         if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
 667                 goto out;
 668         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 669                 goto out;
 670
 671         memslot = &kvm->memslots->memslots[mem->slot];
 672         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 673         npages = mem->memory_size >> PAGE_SHIFT;
 674
 675         r = -EINVAL;
 676         if (npages > KVM_MEM_MAX_NR_PAGES)
 677                 goto out;
 678
 679         if (!npages)
 680                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 681
 682         new = old = *memslot;
 683
 684         new.id = mem->slot;
 685         new.base_gfn = base_gfn;
 686         new.npages = npages;
 687         new.flags = mem->flags;
 688
 689         /* Disallow changing a memory slot's size. */
 690         r = -EINVAL;
 691         if (npages && old.npages && npages != old.npages)
 692                 goto out_free;
 693
 694         /* Check for overlaps */
 695         r = -EEXIST;
 696         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 697                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 698
 699                 if (s == memslot || !s->npages)
 700                         continue;
 701                 if (!((base_gfn + npages <= s->base_gfn) ||
 702                       (base_gfn >= s->base_gfn + s->npages)))
 703                         goto out_free;
 704         }
 705
 706         /* Free page dirty bitmap if unneeded */
 707         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 708                 new.dirty_bitmap = NULL;
 709
 710         r = -ENOMEM;
 711
 712         /* Allocate if a slot is being created */
 713 #ifndef CONFIG_S390
 714         if (npages && !new.rmap) {
 715                 new.rmap = vzalloc(npages * sizeof(*new.rmap));
 716
 717                 if (!new.rmap)
 718                         goto out_free;
 719
 720                 new.user_alloc = user_alloc;
 721                 new.userspace_addr = mem->userspace_addr;
 722         }
 723         if (!npages)
 724                 goto skip_lpage;
 725
 726         for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
 727                 unsigned long ugfn;
 728                 unsigned long j;
 729                 int lpages;
 730                 int level = i + 2;
 731
 732                 /* Avoid unused variable warning if no large pages */
 733                 (void)level;
 734
 735                 if (new.lpage_info[i])
 736                         continue;
 737
 738                 lpages = 1 + ((base_gfn + npages - 1)
 739                              >> KVM_HPAGE_GFN_SHIFT(level));
 740                 lpages -= base_gfn >> KVM_HPAGE_GFN_SHIFT(level);
 741
 742                 new.lpage_info[i] = vzalloc(lpages * sizeof(*new.lpage_info[i]));
 743
 744                 if (!new.lpage_info[i])
 745                         goto out_free;
 746
 747                 if (base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
 748                         new.lpage_info[i][0].write_count = 1;
 749                 if ((base_gfn+npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
 750                         new.lpage_info[i][lpages - 1].write_count = 1;
 751                 ugfn = new.userspace_addr >> PAGE_SHIFT;
 752                 /*
 753                  * If the gfn and userspace address are not aligned wrt each
 754                  * other, or if explicitly asked to, disable large page
 755                  * support for this slot
 756                  */
 757                 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
 758                     !largepages_enabled)
 759                         for (j = 0; j < lpages; ++j)
 760                                 new.lpage_info[i][j].write_count = 1;
 761         }
 762
 763 skip_lpage:
 764
 765         /* Allocate page dirty bitmap if needed */
 766         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 767                 if (kvm_create_dirty_bitmap(&new) < 0)
 768                         goto out_free;
 769                 /* destroy any largepage mappings for dirty tracking */
 770         }
 771 #else  /* not defined CONFIG_S390 */
 772         new.user_alloc = user_alloc;
 773         if (user_alloc)
 774                 new.userspace_addr = mem->userspace_addr;
 775 #endif /* not defined CONFIG_S390 */
 776
 777         if (!npages || base_gfn != old.base_gfn) {
 778                 r = -ENOMEM;
 779                 slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 780                 if (!slots)
 781                         goto out_free;
 782                 memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 783                 if (mem->slot >= slots->nmemslots)
 784                         slots->nmemslots = mem->slot + 1;
 785                 slots->generation++;
 786                 slots->memslots[mem->slot].flags |= KVM_MEMSLOT_INVALID;
 787
 788                 old_memslots = kvm->memslots;
 789                 rcu_assign_pointer(kvm->memslots, slots);
 790                 synchronize_srcu_expedited(&kvm->srcu);
 791                 /* From this point no new shadow pages pointing to a deleted,
 792                  * or moved, memslot will be created.
 793                  *
 794                  * validation of sp->gfn happens in:
 795                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 796                  *      - kvm_is_visible_gfn (mmu_check_roots)
 797                  */
 798                 kvm_arch_flush_shadow(kvm);
 799                 kfree(old_memslots);
 800         }
 801
 802         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 803         if (r)
 804                 goto out_free;
 805
 806         /* map/unmap the pages in iommu page table */
 807         if (npages) {
 808                 r = kvm_iommu_map_pages(kvm, &new);
 809                 if (r)
 810                         goto out_free;
 811         } else
 812                 kvm_iommu_unmap_pages(kvm, &old);
 813
 814         r = -ENOMEM;
 815         slots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 816         if (!slots)
 817                 goto out_free;
 818         memcpy(slots, kvm->memslots, sizeof(struct kvm_memslots));
 819         if (mem->slot >= slots->nmemslots)
 820                 slots->nmemslots = mem->slot + 1;
 821         slots->generation++;
 822
 823         /* actual memory is freed via old in kvm_free_physmem_slot below */
 824         if (!npages) {
 825                 new.rmap = NULL;
 826                 new.dirty_bitmap = NULL;
 827                 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i)
 828                         new.lpage_info[i] = NULL;
 829         }
 830
 831         slots->memslots[mem->slot] = new;
 832         old_memslots = kvm->memslots;
 833         rcu_assign_pointer(kvm->memslots, slots);
 834         synchronize_srcu_expedited(&kvm->srcu);
 835
 836         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 837
 838         /*
 839          * If the new memory slot is created, we need to clear all
 840          * mmio sptes.
 841          */
 842         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 843                 kvm_arch_flush_shadow(kvm);
 844
 845         kvm_free_physmem_slot(&old, &new);
 846         kfree(old_memslots);
 847
 848         return 0;
 849
 850 out_free:
 851         kvm_free_physmem_slot(&new, &old);
 852 out:
 853         return r;
 854
 855 }
 856 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 857
 858 int kvm_set_memory_region(struct kvm *kvm,
 859                           struct kvm_userspace_memory_region *mem,
 860                           int user_alloc)
 861 {
 862         int r;
 863
 864         mutex_lock(&kvm->slots_lock);
 865         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 866         mutex_unlock(&kvm->slots_lock);
 867         return r;
 868 }
 869 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 870
 871 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 872                                    struct
 873                                    kvm_userspace_memory_region *mem,
 874                                    int user_alloc)
 875 {
 876         if (mem->slot >= KVM_MEMORY_SLOTS)
 877                 return -EINVAL;
 878         return kvm_set_memory_region(kvm, mem, user_alloc);
 879 }
 880
 881 int kvm_get_dirty_log(struct kvm *kvm,
 882                         struct kvm_dirty_log *log, int *is_dirty)
 883 {
 884         struct kvm_memory_slot *memslot;
 885         int r, i;
 886         unsigned long n;
 887         unsigned long any = 0;
 888
 889         r = -EINVAL;
 890         if (log->slot >= KVM_MEMORY_SLOTS)
 891                 goto out;
 892
 893         memslot = &kvm->memslots->memslots[log->slot];
 894         r = -ENOENT;
 895         if (!memslot->dirty_bitmap)
 896                 goto out;
 897
 898         n = kvm_dirty_bitmap_bytes(memslot);
 899
 900         for (i = 0; !any && i < n/sizeof(long); ++i)
 901                 any = memslot->dirty_bitmap[i];
 902
 903         r = -EFAULT;
 904         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 905                 goto out;
 906
 907         if (any)
 908                 *is_dirty = 1;
 909
 910         r = 0;
 911 out:
 912         return r;
 913 }
 914
 915 void kvm_disable_largepages(void)
 916 {
 917         largepages_enabled = false;
 918 }
 919 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 920
 921 int is_error_page(struct page *page)
 922 {
 923         return page == bad_page || page == hwpoison_page || page == fault_page;
 924 }
 925 EXPORT_SYMBOL_GPL(is_error_page);
 926
 927 int is_error_pfn(pfn_t pfn)
 928 {
 929         return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
 930 }
 931 EXPORT_SYMBOL_GPL(is_error_pfn);
 932
 933 int is_hwpoison_pfn(pfn_t pfn)
 934 {
 935         return pfn == hwpoison_pfn;
 936 }
 937 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
 938
 939 int is_fault_pfn(pfn_t pfn)
 940 {
 941         return pfn == fault_pfn;
 942 }
 943 EXPORT_SYMBOL_GPL(is_fault_pfn);
 944
 945 int is_noslot_pfn(pfn_t pfn)
 946 {
 947         return pfn == bad_pfn;
 948 }
 949 EXPORT_SYMBOL_GPL(is_noslot_pfn);
 950
 951 int is_invalid_pfn(pfn_t pfn)
 952 {
 953         return pfn == hwpoison_pfn || pfn == fault_pfn;
 954 }
 955 EXPORT_SYMBOL_GPL(is_invalid_pfn);
 956
 957 static inline unsigned long bad_hva(void)
 958 {
 959         return PAGE_OFFSET;
 960 }
 961
 962 int kvm_is_error_hva(unsigned long addr)
 963 {
 964         return addr == bad_hva();
 965 }
 966 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 967
 968 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm_memslots *slots,
 969                                                 gfn_t gfn)
 970 {
 971         int i;
 972
 973         for (i = 0; i < slots->nmemslots; ++i) {
 974                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 975
 976                 if (gfn >= memslot->base_gfn
 977                     && gfn < memslot->base_gfn + memslot->npages)
 978                         return memslot;
 979         }
 980         return NULL;
 981 }
 982
 983 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 984 {
 985         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 986 }
 987 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 988
 989 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 990 {
 991         int i;
 992         struct kvm_memslots *slots = kvm_memslots(kvm);
 993
 994         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 995                 struct kvm_memory_slot *memslot = &slots->memslots[i];
 996
 997                 if (memslot->flags & KVM_MEMSLOT_INVALID)
 998                         continue;
 999
1000                 if (gfn >= memslot->base_gfn
1001                     && gfn < memslot->base_gfn + memslot->npages)
1002                         return 1;
1003         }
1004         return 0;
1005 }
1006 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1007
1008 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1009 {
1010         struct vm_area_struct *vma;
1011         unsigned long addr, size;
1012
1013         size = PAGE_SIZE;
1014
1015         addr = gfn_to_hva(kvm, gfn);
1016         if (kvm_is_error_hva(addr))
1017                 return PAGE_SIZE;
1018
1019         down_read(&current->mm->mmap_sem);
1020         vma = find_vma(current->mm, addr);
1021         if (!vma)
1022                 goto out;
1023
1024         size = vma_kernel_pagesize(vma);
1025
1026 out:
1027         up_read(&current->mm->mmap_sem);
1028
1029         return size;
1030 }
1031
1032 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1033                                      gfn_t *nr_pages)
1034 {
1035         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1036                 return bad_hva();
1037
1038         if (nr_pages)
1039                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
1040
1041         return gfn_to_hva_memslot(slot, gfn);
1042 }
1043
1044 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1045 {
1046         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1047 }
1048 EXPORT_SYMBOL_GPL(gfn_to_hva);
1049
1050 static pfn_t get_fault_pfn(void)
1051 {
1052         get_page(fault_page);
1053         return fault_pfn;
1054 }
1055
1056 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1057         unsigned long start, int write, struct page **page)
1058 {
1059         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1060
1061         if (write)
1062                 flags |= FOLL_WRITE;
1063
1064         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1065 }
1066
1067 static inline int check_user_page_hwpoison(unsigned long addr)
1068 {
1069         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1070
1071         rc = __get_user_pages(current, current->mm, addr, 1,
1072                               flags, NULL, NULL, NULL);
1073         return rc == -EHWPOISON;
1074 }
1075
1076 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1077                         bool *async, bool write_fault, bool *writable)
1078 {
1079         struct page *page[1];
1080         int npages = 0;
1081         pfn_t pfn;
1082
1083         /* we can do it either atomically or asynchronously, not both */
1084         BUG_ON(atomic && async);
1085
1086         BUG_ON(!write_fault && !writable);
1087
1088         if (writable)
1089                 *writable = true;
1090
1091         if (atomic || async)
1092                 npages = __get_user_pages_fast(addr, 1, 1, page);
1093
1094         if (unlikely(npages != 1) && !atomic) {
1095                 might_sleep();
1096
1097                 if (writable)
1098                         *writable = write_fault;
1099
1100                 if (async) {
1101                         down_read(&current->mm->mmap_sem);
1102                         npages = get_user_page_nowait(current, current->mm,
1103                                                      addr, write_fault, page);
1104                         up_read(&current->mm->mmap_sem);
1105                 } else
1106                         npages = get_user_pages_fast(addr, 1, write_fault,
1107                                                      page);
1108
1109                 /* map read fault as writable if possible */
1110                 if (unlikely(!write_fault) && npages == 1) {
1111                         struct page *wpage[1];
1112
1113                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1114                         if (npages == 1) {
1115                                 *writable = true;
1116                                 put_page(page[0]);
1117                                 page[0] = wpage[0];
1118                         }
1119                         npages = 1;
1120                 }
1121         }
1122
1123         if (unlikely(npages != 1)) {
1124                 struct vm_area_struct *vma;
1125
1126                 if (atomic)
1127                         return get_fault_pfn();
1128
1129                 down_read(&current->mm->mmap_sem);
1130                 if (npages == -EHWPOISON ||
1131                         (!async && check_user_page_hwpoison(addr))) {
1132                         up_read(&current->mm->mmap_sem);
1133                         get_page(hwpoison_page);
1134                         return page_to_pfn(hwpoison_page);
1135                 }
1136
1137                 vma = find_vma_intersection(current->mm, addr, addr+1);
1138
1139                 if (vma == NULL)
1140                         pfn = get_fault_pfn();
1141                 else if ((vma->vm_flags & VM_PFNMAP)) {
1142                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1143                                 vma->vm_pgoff;
1144                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1145                 } else {
1146                         if (async && (vma->vm_flags & VM_WRITE))
1147                                 *async = true;
1148                         pfn = get_fault_pfn();
1149                 }
1150                 up_read(&current->mm->mmap_sem);
1151         } else
1152                 pfn = page_to_pfn(page[0]);
1153
1154         return pfn;
1155 }
1156
1157 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1158 {
1159         return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1160 }
1161 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1162
1163 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1164                           bool write_fault, bool *writable)
1165 {
1166         unsigned long addr;
1167
1168         if (async)
1169                 *async = false;
1170
1171         addr = gfn_to_hva(kvm, gfn);
1172         if (kvm_is_error_hva(addr)) {
1173                 get_page(bad_page);
1174                 return page_to_pfn(bad_page);
1175         }
1176
1177         return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1178 }
1179
1180 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1181 {
1182         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1183 }
1184 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1185
1186 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1187                        bool write_fault, bool *writable)
1188 {
1189         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1190 }
1191 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1192
1193 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1194 {
1195         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1196 }
1197 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1198
1199 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1200                       bool *writable)
1201 {
1202         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1203 }
1204 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1205
1206 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1207                          struct kvm_memory_slot *slot, gfn_t gfn)
1208 {
1209         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1210         return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1211 }
1212
1213 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1214                                                                   int nr_pages)
1215 {
1216         unsigned long addr;
1217         gfn_t entry;
1218
1219         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1220         if (kvm_is_error_hva(addr))
1221                 return -1;
1222
1223         if (entry < nr_pages)
1224                 return 0;
1225
1226         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1227 }
1228 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1229
1230 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1231 {
1232         pfn_t pfn;
1233
1234         pfn = gfn_to_pfn(kvm, gfn);
1235         if (!kvm_is_mmio_pfn(pfn))
1236                 return pfn_to_page(pfn);
1237
1238         WARN_ON(kvm_is_mmio_pfn(pfn));
1239
1240         get_page(bad_page);
1241         return bad_page;
1242 }
1243
1244 EXPORT_SYMBOL_GPL(gfn_to_page);
1245
1246 void kvm_release_page_clean(struct page *page)
1247 {
1248         kvm_release_pfn_clean(page_to_pfn(page));
1249 }
1250 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1251
1252 void kvm_release_pfn_clean(pfn_t pfn)
1253 {
1254         if (!kvm_is_mmio_pfn(pfn))
1255                 put_page(pfn_to_page(pfn));
1256 }
1257 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1258
1259 void kvm_release_page_dirty(struct page *page)
1260 {
1261         kvm_release_pfn_dirty(page_to_pfn(page));
1262 }
1263 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1264
1265 void kvm_release_pfn_dirty(pfn_t pfn)
1266 {
1267         kvm_set_pfn_dirty(pfn);
1268         kvm_release_pfn_clean(pfn);
1269 }
1270 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1271
1272 void kvm_set_page_dirty(struct page *page)
1273 {
1274         kvm_set_pfn_dirty(page_to_pfn(page));
1275 }
1276 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1277
1278 void kvm_set_pfn_dirty(pfn_t pfn)
1279 {
1280         if (!kvm_is_mmio_pfn(pfn)) {
1281                 struct page *page = pfn_to_page(pfn);
1282                 if (!PageReserved(page))
1283                         SetPageDirty(page);
1284         }
1285 }
1286 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1287
1288 void kvm_set_pfn_accessed(pfn_t pfn)
1289 {
1290         if (!kvm_is_mmio_pfn(pfn))
1291                 mark_page_accessed(pfn_to_page(pfn));
1292 }
1293 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1294
1295 void kvm_get_pfn(pfn_t pfn)
1296 {
1297         if (!kvm_is_mmio_pfn(pfn))
1298                 get_page(pfn_to_page(pfn));
1299 }
1300 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1301
1302 static int next_segment(unsigned long len, int offset)
1303 {
1304         if (len > PAGE_SIZE - offset)
1305                 return PAGE_SIZE - offset;
1306         else
1307                 return len;
1308 }
1309
1310 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1311                         int len)
1312 {
1313         int r;
1314         unsigned long addr;
1315
1316         addr = gfn_to_hva(kvm, gfn);
1317         if (kvm_is_error_hva(addr))
1318                 return -EFAULT;
1319         r = __copy_from_user(data, (void __user *)addr + offset, len);
1320         if (r)
1321                 return -EFAULT;
1322         return 0;
1323 }
1324 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1325
1326 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1327 {
1328         gfn_t gfn = gpa >> PAGE_SHIFT;
1329         int seg;
1330         int offset = offset_in_page(gpa);
1331         int ret;
1332
1333         while ((seg = next_segment(len, offset)) != 0) {
1334                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1335                 if (ret < 0)
1336                         return ret;
1337                 offset = 0;
1338                 len -= seg;
1339                 data += seg;
1340                 ++gfn;
1341         }
1342         return 0;
1343 }
1344 EXPORT_SYMBOL_GPL(kvm_read_guest);
1345
1346 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1347                           unsigned long len)
1348 {
1349         int r;
1350         unsigned long addr;
1351         gfn_t gfn = gpa >> PAGE_SHIFT;
1352         int offset = offset_in_page(gpa);
1353
1354         addr = gfn_to_hva(kvm, gfn);
1355         if (kvm_is_error_hva(addr))
1356                 return -EFAULT;
1357         pagefault_disable();
1358         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1359         pagefault_enable();
1360         if (r)
1361                 return -EFAULT;
1362         return 0;
1363 }
1364 EXPORT_SYMBOL(kvm_read_guest_atomic);
1365
1366 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1367                          int offset, int len)
1368 {
1369         int r;
1370         unsigned long addr;
1371
1372         addr = gfn_to_hva(kvm, gfn);
1373         if (kvm_is_error_hva(addr))
1374                 return -EFAULT;
1375         r = __copy_to_user((void __user *)addr + offset, data, len);
1376         if (r)
1377                 return -EFAULT;
1378         mark_page_dirty(kvm, gfn);
1379         return 0;
1380 }
1381 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1382
1383 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1384                     unsigned long len)
1385 {
1386         gfn_t gfn = gpa >> PAGE_SHIFT;
1387         int seg;
1388         int offset = offset_in_page(gpa);
1389         int ret;
1390
1391         while ((seg = next_segment(len, offset)) != 0) {
1392                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1393                 if (ret < 0)
1394                         return ret;
1395                 offset = 0;
1396                 len -= seg;
1397                 data += seg;
1398                 ++gfn;
1399         }
1400         return 0;
1401 }
1402
1403 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1404                               gpa_t gpa, unsigned long len)
1405 {
1406         struct kvm_memslots *slots = kvm_memslots(kvm);
1407         int offset = offset_in_page(gpa);
1408         gfn_t start_gfn = gpa >> PAGE_SHIFT;
1409         gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1410         gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1411         gfn_t nr_pages_avail;
1412
1413         ghc->gpa = gpa;
1414         ghc->generation = slots->generation;
1415         ghc->len = len;
1416         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1417         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, &nr_pages_avail);
1418         if (!kvm_is_error_hva(ghc->hva) && nr_pages_avail >= nr_pages_needed) {
1419                 ghc->hva += offset;
1420         } else {
1421                 /*
1422                  * If the requested region crosses two memslots, we still
1423                  * verify that the entire region is valid here.
1424                  */
1425                 while (start_gfn <= end_gfn) {
1426                         ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1427                         ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1428                                                    &nr_pages_avail);
1429                         if (kvm_is_error_hva(ghc->hva))
1430                                 return -EFAULT;
1431                         start_gfn += nr_pages_avail;
1432                 }
1433                 /* Use the slow path for cross page reads and writes. */
1434                 ghc->memslot = NULL;
1435         }
1436         return 0;
1437 }
1438 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1439
1440 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1441                            void *data, unsigned long len)
1442 {
1443         struct kvm_memslots *slots = kvm_memslots(kvm);
1444         int r;
1445
1446         BUG_ON(len > ghc->len);
1447
1448         if (slots->generation != ghc->generation)
1449                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1450
1451         if (unlikely(!ghc->memslot))
1452                 return kvm_write_guest(kvm, ghc->gpa, data, len);
1453
1454         if (kvm_is_error_hva(ghc->hva))
1455                 return -EFAULT;
1456
1457         r = __copy_to_user((void __user *)ghc->hva, data, len);
1458         if (r)
1459                 return -EFAULT;
1460         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1461
1462         return 0;
1463 }
1464 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1465
1466 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1467                            void *data, unsigned long len)
1468 {
1469         struct kvm_memslots *slots = kvm_memslots(kvm);
1470         int r;
1471
1472         BUG_ON(len > ghc->len);
1473
1474         if (slots->generation != ghc->generation)
1475                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1476
1477         if (unlikely(!ghc->memslot))
1478                 return kvm_read_guest(kvm, ghc->gpa, data, len);
1479
1480         if (kvm_is_error_hva(ghc->hva))
1481                 return -EFAULT;
1482
1483         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1484         if (r)
1485                 return -EFAULT;
1486
1487         return 0;
1488 }
1489 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1490
1491 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1492 {
1493         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1494                                     offset, len);
1495 }
1496 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1497
1498 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1499 {
1500         gfn_t gfn = gpa >> PAGE_SHIFT;
1501         int seg;
1502         int offset = offset_in_page(gpa);
1503         int ret;
1504
1505         while ((seg = next_segment(len, offset)) != 0) {
1506                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1507                 if (ret < 0)
1508                         return ret;
1509                 offset = 0;
1510                 len -= seg;
1511                 ++gfn;
1512         }
1513         return 0;
1514 }
1515 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1516
1517 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1518                              gfn_t gfn)
1519 {
1520         if (memslot && memslot->dirty_bitmap) {
1521                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1522
1523                 __set_bit_le(rel_gfn, memslot->dirty_bitmap);
1524         }
1525 }
1526
1527 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1528 {
1529         struct kvm_memory_slot *memslot;
1530
1531         memslot = gfn_to_memslot(kvm, gfn);
1532         mark_page_dirty_in_slot(kvm, memslot, gfn);
1533 }
1534
1535 /*
1536  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1537  */
1538 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1539 {
1540         DEFINE_WAIT(wait);
1541
1542         for (;;) {
1543                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1544
1545                 if (kvm_arch_vcpu_runnable(vcpu)) {
1546                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1547                         break;
1548                 }
1549                 if (kvm_cpu_has_pending_timer(vcpu))
1550                         break;
1551                 if (signal_pending(current))
1552                         break;
1553
1554                 schedule();
1555         }
1556
1557         finish_wait(&vcpu->wq, &wait);
1558 }
1559
1560 void kvm_resched(struct kvm_vcpu *vcpu)
1561 {
1562         if (!need_resched())
1563                 return;
1564         cond_resched();
1565 }
1566 EXPORT_SYMBOL_GPL(kvm_resched);
1567
1568 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1569 {
1570         struct kvm *kvm = me->kvm;
1571         struct kvm_vcpu *vcpu;
1572         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1573         int yielded = 0;
1574         int pass;
1575         int i;
1576
1577         /*
1578          * We boost the priority of a VCPU that is runnable but not
1579          * currently running, because it got preempted by something
1580          * else and called schedule in __vcpu_run.  Hopefully that
1581          * VCPU is holding the lock that we need and will release it.
1582          * We approximate round-robin by starting at the last boosted VCPU.
1583          */
1584         for (pass = 0; pass < 2 && !yielded; pass++) {
1585                 kvm_for_each_vcpu(i, vcpu, kvm) {
1586                         struct task_struct *task = NULL;
1587                         struct pid *pid;
1588                         if (!pass && i < last_boosted_vcpu) {
1589                                 i = last_boosted_vcpu;
1590                                 continue;
1591                         } else if (pass && i > last_boosted_vcpu)
1592                                 break;
1593                         if (vcpu == me)
1594                                 continue;
1595                         if (waitqueue_active(&vcpu->wq))
1596                                 continue;
1597                         rcu_read_lock();
1598                         pid = rcu_dereference(vcpu->pid);
1599                         if (pid)
1600                                 task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1601                         rcu_read_unlock();
1602                         if (!task)
1603                                 continue;
1604                         if (task->flags & PF_VCPU) {
1605                                 put_task_struct(task);
1606                                 continue;
1607                         }
1608                         if (yield_to(task, 1)) {
1609                                 put_task_struct(task);
1610                                 kvm->last_boosted_vcpu = i;
1611                                 yielded = 1;
1612                                 break;
1613                         }
1614                         put_task_struct(task);
1615                 }
1616         }
1617 }
1618 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1619
1620 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1621 {
1622         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1623         struct page *page;
1624
1625         if (vmf->pgoff == 0)
1626                 page = virt_to_page(vcpu->run);
1627 #ifdef CONFIG_X86
1628         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1629                 page = virt_to_page(vcpu->arch.pio_data);
1630 #endif
1631 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1632         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1633                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1634 #endif
1635         else
1636                 return VM_FAULT_SIGBUS;
1637         get_page(page);
1638         vmf->page = page;
1639         return 0;
1640 }
1641
1642 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1643         .fault = kvm_vcpu_fault,
1644 };
1645
1646 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1647 {
1648         vma->vm_ops = &kvm_vcpu_vm_ops;
1649         return 0;
1650 }
1651
1652 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1653 {
1654         struct kvm_vcpu *vcpu = filp->private_data;
1655
1656         kvm_put_kvm(vcpu->kvm);
1657         return 0;
1658 }
1659
1660 static struct file_operations kvm_vcpu_fops = {
1661         .release        = kvm_vcpu_release,
1662         .unlocked_ioctl = kvm_vcpu_ioctl,
1663 #ifdef CONFIG_COMPAT
1664         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1665 #endif
1666         .mmap           = kvm_vcpu_mmap,
1667         .llseek         = noop_llseek,
1668 };
1669
1670 /*
1671  * Allocates an inode for the vcpu.
1672  */
1673 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1674 {
1675         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1676 }
1677
1678 /*
1679  * Creates some virtual cpus.  Good luck creating more than one.
1680  */
1681 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1682 {
1683         int r;
1684         struct kvm_vcpu *vcpu, *v;
1685
1686         if (id >= KVM_MAX_VCPUS)
1687                 return -EINVAL;
1688
1689         vcpu = kvm_arch_vcpu_create(kvm, id);
1690         if (IS_ERR(vcpu))
1691                 return PTR_ERR(vcpu);
1692
1693         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1694
1695         r = kvm_arch_vcpu_setup(vcpu);
1696         if (r)
1697                 goto vcpu_destroy;
1698
1699         mutex_lock(&kvm->lock);
1700         if (!kvm_vcpu_compatible(vcpu)) {
1701                 r = -EINVAL;
1702                 goto unlock_vcpu_destroy;
1703         }
1704         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1705                 r = -EINVAL;
1706                 goto unlock_vcpu_destroy;
1707         }
1708
1709         kvm_for_each_vcpu(r, v, kvm)
1710                 if (v->vcpu_id == id) {
1711                         r = -EEXIST;
1712                         goto unlock_vcpu_destroy;
1713                 }
1714
1715         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1716
1717         /* Now it's all set up, let userspace reach it */
1718         kvm_get_kvm(kvm);
1719         r = create_vcpu_fd(vcpu);
1720         if (r < 0) {
1721                 kvm_put_kvm(kvm);
1722                 goto unlock_vcpu_destroy;
1723         }
1724
1725         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1726         smp_wmb();
1727         atomic_inc(&kvm->online_vcpus);
1728
1729 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
1730         if (kvm->bsp_vcpu_id == id)
1731                 kvm->bsp_vcpu = vcpu;
1732 #endif
1733         mutex_unlock(&kvm->lock);
1734         return r;
1735
1736 unlock_vcpu_destroy:
1737         mutex_unlock(&kvm->lock);
1738 vcpu_destroy:
1739         kvm_arch_vcpu_destroy(vcpu);
1740         return r;
1741 }
1742
1743 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1744 {
1745         if (sigset) {
1746                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1747                 vcpu->sigset_active = 1;
1748                 vcpu->sigset = *sigset;
1749         } else
1750                 vcpu->sigset_active = 0;
1751         return 0;
1752 }
1753
1754 static long kvm_vcpu_ioctl(struct file *filp,
1755                            unsigned int ioctl, unsigned long arg)
1756 {
1757         struct kvm_vcpu *vcpu = filp->private_data;
1758         void __user *argp = (void __user *)arg;
1759         int r;
1760         struct kvm_fpu *fpu = NULL;
1761         struct kvm_sregs *kvm_sregs = NULL;
1762
1763         if (vcpu->kvm->mm != current->mm)
1764                 return -EIO;
1765
1766 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1767         /*
1768          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1769          * so vcpu_load() would break it.
1770          */
1771         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1772                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1773 #endif
1774
1775
1776         vcpu_load(vcpu);
1777         switch (ioctl) {
1778         case KVM_RUN:
1779                 r = -EINVAL;
1780                 if (arg)
1781                         goto out;
1782                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1783                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1784                 break;
1785         case KVM_GET_REGS: {
1786                 struct kvm_regs *kvm_regs;
1787
1788                 r = -ENOMEM;
1789                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1790                 if (!kvm_regs)
1791                         goto out;
1792                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1793                 if (r)
1794                         goto out_free1;
1795                 r = -EFAULT;
1796                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1797                         goto out_free1;
1798                 r = 0;
1799 out_free1:
1800                 kfree(kvm_regs);
1801                 break;
1802         }
1803         case KVM_SET_REGS: {
1804                 struct kvm_regs *kvm_regs;
1805
1806                 r = -ENOMEM;
1807                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1808                 if (!kvm_regs)
1809                         goto out;
1810                 r = -EFAULT;
1811                 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1812                         goto out_free2;
1813                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1814                 if (r)
1815                         goto out_free2;
1816                 r = 0;
1817 out_free2:
1818                 kfree(kvm_regs);
1819                 break;
1820         }
1821         case KVM_GET_SREGS: {
1822                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1823                 r = -ENOMEM;
1824                 if (!kvm_sregs)
1825                         goto out;
1826                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1827                 if (r)
1828                         goto out;
1829                 r = -EFAULT;
1830                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1831                         goto out;
1832                 r = 0;
1833                 break;
1834         }
1835         case KVM_SET_SREGS: {
1836                 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1837                 r = -ENOMEM;
1838                 if (!kvm_sregs)
1839                         goto out;
1840                 r = -EFAULT;
1841                 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1842                         goto out;
1843                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1844                 if (r)
1845                         goto out;
1846                 r = 0;
1847                 break;
1848         }
1849         case KVM_GET_MP_STATE: {
1850                 struct kvm_mp_state mp_state;
1851
1852                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1853                 if (r)
1854                         goto out;
1855                 r = -EFAULT;
1856                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1857                         goto out;
1858                 r = 0;
1859                 break;
1860         }
1861         case KVM_SET_MP_STATE: {
1862                 struct kvm_mp_state mp_state;
1863
1864                 r = -EFAULT;
1865                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1866                         goto out;
1867                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1868                 if (r)
1869                         goto out;
1870                 r = 0;
1871                 break;
1872         }
1873         case KVM_TRANSLATE: {
1874                 struct kvm_translation tr;
1875
1876                 r = -EFAULT;
1877                 if (copy_from_user(&tr, argp, sizeof tr))
1878                         goto out;
1879                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1880                 if (r)
1881                         goto out;
1882                 r = -EFAULT;
1883                 if (copy_to_user(argp, &tr, sizeof tr))
1884                         goto out;
1885                 r = 0;
1886                 break;
1887         }
1888         case KVM_SET_GUEST_DEBUG: {
1889                 struct kvm_guest_debug dbg;
1890
1891                 r = -EFAULT;
1892                 if (copy_from_user(&dbg, argp, sizeof dbg))
1893                         goto out;
1894                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1895                 if (r)
1896                         goto out;
1897                 r = 0;
1898                 break;
1899         }
1900         case KVM_SET_SIGNAL_MASK: {
1901                 struct kvm_signal_mask __user *sigmask_arg = argp;
1902                 struct kvm_signal_mask kvm_sigmask;
1903                 sigset_t sigset, *p;
1904
1905                 p = NULL;
1906                 if (argp) {
1907                         r = -EFAULT;
1908                         if (copy_from_user(&kvm_sigmask, argp,
1909                                            sizeof kvm_sigmask))
1910                                 goto out;
1911                         r = -EINVAL;
1912                         if (kvm_sigmask.len != sizeof sigset)
1913                                 goto out;
1914                         r = -EFAULT;
1915                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1916                                            sizeof sigset))
1917                                 goto out;
1918                         p = &sigset;
1919                 }
1920                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1921                 break;
1922         }
1923         case KVM_GET_FPU: {
1924                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1925                 r = -ENOMEM;
1926                 if (!fpu)
1927                         goto out;
1928                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1929                 if (r)
1930                         goto out;
1931                 r = -EFAULT;
1932                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1933                         goto out;
1934                 r = 0;
1935                 break;
1936         }
1937         case KVM_SET_FPU: {
1938                 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1939                 r = -ENOMEM;
1940                 if (!fpu)
1941                         goto out;
1942                 r = -EFAULT;
1943                 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1944                         goto out;
1945                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1946                 if (r)
1947                         goto out;
1948                 r = 0;
1949                 break;
1950         }
1951         default:
1952                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1953         }
1954 out:
1955         vcpu_put(vcpu);
1956         kfree(fpu);
1957         kfree(kvm_sregs);
1958         return r;
1959 }
1960
1961 #ifdef CONFIG_COMPAT
1962 static long kvm_vcpu_compat_ioctl(struct file *filp,
1963                                   unsigned int ioctl, unsigned long arg)
1964 {
1965         struct kvm_vcpu *vcpu = filp->private_data;
1966         void __user *argp = compat_ptr(arg);
1967         int r;
1968
1969         if (vcpu->kvm->mm != current->mm)
1970                 return -EIO;
1971
1972         switch (ioctl) {
1973         case KVM_SET_SIGNAL_MASK: {
1974                 struct kvm_signal_mask __user *sigmask_arg = argp;
1975                 struct kvm_signal_mask kvm_sigmask;
1976                 compat_sigset_t csigset;
1977                 sigset_t sigset;
1978
1979                 if (argp) {
1980                         r = -EFAULT;
1981                         if (copy_from_user(&kvm_sigmask, argp,
1982                                            sizeof kvm_sigmask))
1983                                 goto out;
1984                         r = -EINVAL;
1985                         if (kvm_sigmask.len != sizeof csigset)
1986                                 goto out;
1987                         r = -EFAULT;
1988                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1989                                            sizeof csigset))
1990                                 goto out;
1991                 }
1992                 sigset_from_compat(&sigset, &csigset);
1993                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1994                 break;
1995         }
1996         default:
1997                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1998         }
1999
2000 out:
2001         return r;
2002 }
2003 #endif
2004
2005 static long kvm_vm_ioctl(struct file *filp,
2006                            unsigned int ioctl, unsigned long arg)
2007 {
2008         struct kvm *kvm = filp->private_data;
2009         void __user *argp = (void __user *)arg;
2010         int r;
2011
2012         if (kvm->mm != current->mm)
2013                 return -EIO;
2014         switch (ioctl) {
2015         case KVM_CREATE_VCPU:
2016                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2017                 if (r < 0)
2018                         goto out;
2019                 break;
2020         case KVM_SET_USER_MEMORY_REGION: {
2021                 struct kvm_userspace_memory_region kvm_userspace_mem;
2022
2023                 r = -EFAULT;
2024                 if (copy_from_user(&kvm_userspace_mem, argp,
2025                                                 sizeof kvm_userspace_mem))
2026                         goto out;
2027
2028                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2029                 if (r)
2030                         goto out;
2031                 break;
2032         }
2033         case KVM_GET_DIRTY_LOG: {
2034                 struct kvm_dirty_log log;
2035
2036                 r = -EFAULT;
2037                 if (copy_from_user(&log, argp, sizeof log))
2038                         goto out;
2039                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2040                 if (r)
2041                         goto out;
2042                 break;
2043         }
2044 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2045         case KVM_REGISTER_COALESCED_MMIO: {
2046                 struct kvm_coalesced_mmio_zone zone;
2047                 r = -EFAULT;
2048                 if (copy_from_user(&zone, argp, sizeof zone))
2049                         goto out;
2050                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2051                 if (r)
2052                         goto out;
2053                 r = 0;
2054                 break;
2055         }
2056         case KVM_UNREGISTER_COALESCED_MMIO: {
2057                 struct kvm_coalesced_mmio_zone zone;
2058                 r = -EFAULT;
2059                 if (copy_from_user(&zone, argp, sizeof zone))
2060                         goto out;
2061                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2062                 if (r)
2063                         goto out;
2064                 r = 0;
2065                 break;
2066         }
2067 #endif
2068         case KVM_IRQFD: {
2069                 struct kvm_irqfd data;
2070
2071                 r = -EFAULT;
2072                 if (copy_from_user(&data, argp, sizeof data))
2073                         goto out;
2074                 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2075                 break;
2076         }
2077         case KVM_IOEVENTFD: {
2078                 struct kvm_ioeventfd data;
2079
2080                 r = -EFAULT;
2081                 if (copy_from_user(&data, argp, sizeof data))
2082                         goto out;
2083                 r = kvm_ioeventfd(kvm, &data);
2084                 break;
2085         }
2086 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2087         case KVM_SET_BOOT_CPU_ID:
2088                 r = 0;
2089                 mutex_lock(&kvm->lock);
2090                 if (atomic_read(&kvm->online_vcpus) != 0)
2091                         r = -EBUSY;
2092                 else
2093                         kvm->bsp_vcpu_id = arg;
2094                 mutex_unlock(&kvm->lock);
2095                 break;
2096 #endif
2097         default:
2098                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2099                 if (r == -ENOTTY)
2100                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2101         }
2102 out:
2103         return r;
2104 }
2105
2106 #ifdef CONFIG_COMPAT
2107 struct compat_kvm_dirty_log {
2108         __u32 slot;
2109         __u32 padding1;
2110         union {
2111                 compat_uptr_t dirty_bitmap; /* one bit per page */
2112                 __u64 padding2;
2113         };
2114 };
2115
2116 static long kvm_vm_compat_ioctl(struct file *filp,
2117                            unsigned int ioctl, unsigned long arg)
2118 {
2119         struct kvm *kvm = filp->private_data;
2120         int r;
2121
2122         if (kvm->mm != current->mm)
2123                 return -EIO;
2124         switch (ioctl) {
2125         case KVM_GET_DIRTY_LOG: {
2126                 struct compat_kvm_dirty_log compat_log;
2127                 struct kvm_dirty_log log;
2128
2129                 r = -EFAULT;
2130                 if (copy_from_user(&compat_log, (void __user *)arg,
2131                                    sizeof(compat_log)))
2132                         goto out;
2133                 log.slot         = compat_log.slot;
2134                 log.padding1     = compat_log.padding1;
2135                 log.padding2     = compat_log.padding2;
2136                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2137
2138                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2139                 if (r)
2140                         goto out;
2141                 break;
2142         }
2143         default:
2144                 r = kvm_vm_ioctl(filp, ioctl, arg);
2145         }
2146
2147 out:
2148         return r;
2149 }
2150 #endif
2151
2152 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2153 {
2154         struct page *page[1];
2155         unsigned long addr;
2156         int npages;
2157         gfn_t gfn = vmf->pgoff;
2158         struct kvm *kvm = vma->vm_file->private_data;
2159
2160         addr = gfn_to_hva(kvm, gfn);
2161         if (kvm_is_error_hva(addr))
2162                 return VM_FAULT_SIGBUS;
2163
2164         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2165                                 NULL);
2166         if (unlikely(npages != 1))
2167                 return VM_FAULT_SIGBUS;
2168
2169         vmf->page = page[0];
2170         return 0;
2171 }
2172
2173 static const struct vm_operations_struct kvm_vm_vm_ops = {
2174         .fault = kvm_vm_fault,
2175 };
2176
2177 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2178 {
2179         vma->vm_ops = &kvm_vm_vm_ops;
2180         return 0;
2181 }
2182
2183 static struct file_operations kvm_vm_fops = {
2184         .release        = kvm_vm_release,
2185         .unlocked_ioctl = kvm_vm_ioctl,
2186 #ifdef CONFIG_COMPAT
2187         .compat_ioctl   = kvm_vm_compat_ioctl,
2188 #endif
2189         .mmap           = kvm_vm_mmap,
2190         .llseek         = noop_llseek,
2191 };
2192
2193 static int kvm_dev_ioctl_create_vm(void)
2194 {
2195         int r;
2196         struct kvm *kvm;
2197
2198         kvm = kvm_create_vm();
2199         if (IS_ERR(kvm))
2200                 return PTR_ERR(kvm);
2201 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2202         r = kvm_coalesced_mmio_init(kvm);
2203         if (r < 0) {
2204                 kvm_put_kvm(kvm);
2205                 return r;
2206         }
2207 #endif
2208         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2209         if (r < 0)
2210                 kvm_put_kvm(kvm);
2211
2212         return r;
2213 }
2214
2215 static long kvm_dev_ioctl_check_extension_generic(long arg)
2216 {
2217         switch (arg) {
2218         case KVM_CAP_USER_MEMORY:
2219         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2220         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2221 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2222         case KVM_CAP_SET_BOOT_CPU_ID:
2223 #endif
2224         case KVM_CAP_INTERNAL_ERROR_DATA:
2225                 return 1;
2226 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2227         case KVM_CAP_IRQ_ROUTING:
2228                 return KVM_MAX_IRQ_ROUTES;
2229 #endif
2230         default:
2231                 break;
2232         }
2233         return kvm_dev_ioctl_check_extension(arg);
2234 }
2235
2236 static long kvm_dev_ioctl(struct file *filp,
2237                           unsigned int ioctl, unsigned long arg)
2238 {
2239         long r = -EINVAL;
2240
2241         switch (ioctl) {
2242         case KVM_GET_API_VERSION:
2243                 r = -EINVAL;
2244                 if (arg)
2245                         goto out;
2246                 r = KVM_API_VERSION;
2247                 break;
2248         case KVM_CREATE_VM:
2249                 r = -EINVAL;
2250                 if (arg)
2251                         goto out;
2252                 r = kvm_dev_ioctl_create_vm();
2253                 break;
2254         case KVM_CHECK_EXTENSION:
2255                 r = kvm_dev_ioctl_check_extension_generic(arg);
2256                 break;
2257         case KVM_GET_VCPU_MMAP_SIZE:
2258                 r = -EINVAL;
2259                 if (arg)
2260                         goto out;
2261                 r = PAGE_SIZE;     /* struct kvm_run */
2262 #ifdef CONFIG_X86
2263                 r += PAGE_SIZE;    /* pio data page */
2264 #endif
2265 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2266                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2267 #endif
2268                 break;
2269         case KVM_TRACE_ENABLE:
2270         case KVM_TRACE_PAUSE:
2271         case KVM_TRACE_DISABLE:
2272                 r = -EOPNOTSUPP;
2273                 break;
2274         default:
2275                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2276         }
2277 out:
2278         return r;
2279 }
2280
2281 static struct file_operations kvm_chardev_ops = {
2282         .unlocked_ioctl = kvm_dev_ioctl,
2283         .compat_ioctl   = kvm_dev_ioctl,
2284         .llseek         = noop_llseek,
2285 };
2286
2287 static struct miscdevice kvm_dev = {
2288         KVM_MINOR,
2289         "kvm",
2290         &kvm_chardev_ops,
2291 };
2292
2293 static void hardware_enable_nolock(void *junk)
2294 {
2295         int cpu = raw_smp_processor_id();
2296         int r;
2297
2298         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2299                 return;
2300
2301         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2302
2303         r = kvm_arch_hardware_enable(NULL);
2304
2305         if (r) {
2306                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2307                 atomic_inc(&hardware_enable_failed);
2308                 printk(KERN_INFO "kvm: enabling virtualization on "
2309                                  "CPU%d failed\n", cpu);
2310         }
2311 }
2312
2313 static void hardware_enable(void *junk)
2314 {
2315         raw_spin_lock(&kvm_lock);
2316         hardware_enable_nolock(junk);
2317         raw_spin_unlock(&kvm_lock);
2318 }
2319
2320 static void hardware_disable_nolock(void *junk)
2321 {
2322         int cpu = raw_smp_processor_id();
2323
2324         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2325                 return;
2326         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2327         kvm_arch_hardware_disable(NULL);
2328 }
2329
2330 static void hardware_disable(void *junk)
2331 {
2332         raw_spin_lock(&kvm_lock);
2333         hardware_disable_nolock(junk);
2334         raw_spin_unlock(&kvm_lock);
2335 }
2336
2337 static void hardware_disable_all_nolock(void)
2338 {
2339         BUG_ON(!kvm_usage_count);
2340
2341         kvm_usage_count--;
2342         if (!kvm_usage_count)
2343                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2344 }
2345
2346 static void hardware_disable_all(void)
2347 {
2348         raw_spin_lock(&kvm_lock);
2349         hardware_disable_all_nolock();
2350         raw_spin_unlock(&kvm_lock);
2351 }
2352
2353 static int hardware_enable_all(void)
2354 {
2355         int r = 0;
2356
2357         raw_spin_lock(&kvm_lock);
2358
2359         kvm_usage_count++;
2360         if (kvm_usage_count == 1) {
2361                 atomic_set(&hardware_enable_failed, 0);
2362                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2363
2364                 if (atomic_read(&hardware_enable_failed)) {
2365                         hardware_disable_all_nolock();
2366                         r = -EBUSY;
2367                 }
2368         }
2369
2370         raw_spin_unlock(&kvm_lock);
2371
2372         return r;
2373 }
2374
2375 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2376                            void *v)
2377 {
2378         int cpu = (long)v;
2379
2380         if (!kvm_usage_count)
2381                 return NOTIFY_OK;
2382
2383         val &= ~CPU_TASKS_FROZEN;
2384         switch (val) {
2385         case CPU_DYING:
2386                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2387                        cpu);
2388                 hardware_disable(NULL);
2389                 break;
2390         case CPU_STARTING:
2391                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2392                        cpu);
2393                 hardware_enable(NULL);
2394                 break;
2395         }
2396         return NOTIFY_OK;
2397 }
2398
2399
2400 asmlinkage void kvm_spurious_fault(void)
2401 {
2402         /* Fault while not rebooting.  We want the trace. */
2403         BUG();
2404 }
2405 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2406
2407 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2408                       void *v)
2409 {
2410         /*
2411          * Some (well, at least mine) BIOSes hang on reboot if
2412          * in vmx root mode.
2413          *
2414          * And Intel TXT required VMX off for all cpu when system shutdown.
2415          */
2416         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2417         kvm_rebooting = true;
2418         on_each_cpu(hardware_disable_nolock, NULL, 1);
2419         return NOTIFY_OK;
2420 }
2421
2422 static struct notifier_block kvm_reboot_notifier = {
2423         .notifier_call = kvm_reboot,
2424         .priority = 0,
2425 };
2426
2427 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2428 {
2429         int i;
2430
2431         for (i = 0; i < bus->dev_count; i++) {
2432                 struct kvm_io_device *pos = bus->range[i].dev;
2433
2434                 kvm_iodevice_destructor(pos);
2435         }
2436         kfree(bus);
2437 }
2438
2439 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2440 {
2441         const struct kvm_io_range *r1 = p1;
2442         const struct kvm_io_range *r2 = p2;
2443
2444         if (r1->addr < r2->addr)
2445                 return -1;
2446         if (r1->addr + r1->len > r2->addr + r2->len)
2447                 return 1;
2448         return 0;
2449 }
2450
2451 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2452                           gpa_t addr, int len)
2453 {
2454         if (bus->dev_count == NR_IOBUS_DEVS)
2455                 return -ENOSPC;
2456
2457         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2458                 .addr = addr,
2459                 .len = len,
2460                 .dev = dev,
2461         };
2462
2463         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2464                 kvm_io_bus_sort_cmp, NULL);
2465
2466         return 0;
2467 }
2468
2469 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2470                              gpa_t addr, int len)
2471 {
2472         struct kvm_io_range *range, key;
2473         int off;
2474
2475         key = (struct kvm_io_range) {
2476                 .addr = addr,
2477                 .len = len,
2478         };
2479
2480         range = bsearch(&key, bus->range, bus->dev_count,
2481                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2482         if (range == NULL)
2483                 return -ENOENT;
2484
2485         off = range - bus->range;
2486
2487         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2488                 off--;
2489
2490         return off;
2491 }
2492
2493 /* kvm_io_bus_write - called under kvm->slots_lock */
2494 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2495                      int len, const void *val)
2496 {
2497         int idx;
2498         struct kvm_io_bus *bus;
2499         struct kvm_io_range range;
2500
2501         range = (struct kvm_io_range) {
2502                 .addr = addr,
2503                 .len = len,
2504         };
2505
2506         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2507         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2508         if (idx < 0)
2509                 return -EOPNOTSUPP;
2510
2511         while (idx < bus->dev_count &&
2512                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2513                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2514                         return 0;
2515                 idx++;
2516         }
2517
2518         return -EOPNOTSUPP;
2519 }
2520
2521 /* kvm_io_bus_read - called under kvm->slots_lock */
2522 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2523                     int len, void *val)
2524 {
2525         int idx;
2526         struct kvm_io_bus *bus;
2527         struct kvm_io_range range;
2528
2529         range = (struct kvm_io_range) {
2530                 .addr = addr,
2531                 .len = len,
2532         };
2533
2534         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2535         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2536         if (idx < 0)
2537                 return -EOPNOTSUPP;
2538
2539         while (idx < bus->dev_count &&
2540                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2541                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2542                         return 0;
2543                 idx++;
2544         }
2545
2546         return -EOPNOTSUPP;
2547 }
2548
2549 /* Caller must hold slots_lock. */
2550 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2551                             int len, struct kvm_io_device *dev)
2552 {
2553         struct kvm_io_bus *new_bus, *bus;
2554
2555         bus = kvm->buses[bus_idx];
2556         if (bus->dev_count > NR_IOBUS_DEVS-1)
2557                 return -ENOSPC;
2558
2559         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2560         if (!new_bus)
2561                 return -ENOMEM;
2562         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2563         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2564         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2565         synchronize_srcu_expedited(&kvm->srcu);
2566         kfree(bus);
2567
2568         return 0;
2569 }
2570
2571 /* Caller must hold slots_lock. */
2572 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2573                               struct kvm_io_device *dev)
2574 {
2575         int i, r;
2576         struct kvm_io_bus *new_bus, *bus;
2577
2578         new_bus = kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL);
2579         if (!new_bus)
2580                 return -ENOMEM;
2581
2582         bus = kvm->buses[bus_idx];
2583         memcpy(new_bus, bus, sizeof(struct kvm_io_bus));
2584
2585         r = -ENOENT;
2586         for (i = 0; i < new_bus->dev_count; i++)
2587                 if (new_bus->range[i].dev == dev) {
2588                         r = 0;
2589                         new_bus->dev_count--;
2590                         new_bus->range[i] = new_bus->range[new_bus->dev_count];
2591                         sort(new_bus->range, new_bus->dev_count,
2592                              sizeof(struct kvm_io_range),
2593                              kvm_io_bus_sort_cmp, NULL);
2594                         break;
2595                 }
2596
2597         if (r) {
2598                 kfree(new_bus);
2599                 return r;
2600         }
2601
2602         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2603         synchronize_srcu_expedited(&kvm->srcu);
2604         kfree(bus);
2605         return r;
2606 }
2607
2608 static struct notifier_block kvm_cpu_notifier = {
2609         .notifier_call = kvm_cpu_hotplug,
2610 };
2611
2612 static int vm_stat_get(void *_offset, u64 *val)
2613 {
2614         unsigned offset = (long)_offset;
2615         struct kvm *kvm;
2616
2617         *val = 0;
2618         raw_spin_lock(&kvm_lock);
2619         list_for_each_entry(kvm, &vm_list, vm_list)
2620                 *val += *(u32 *)((void *)kvm + offset);
2621         raw_spin_unlock(&kvm_lock);
2622         return 0;
2623 }
2624
2625 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2626
2627 static int vcpu_stat_get(void *_offset, u64 *val)
2628 {
2629         unsigned offset = (long)_offset;
2630         struct kvm *kvm;
2631         struct kvm_vcpu *vcpu;
2632         int i;
2633
2634         *val = 0;
2635         raw_spin_lock(&kvm_lock);
2636         list_for_each_entry(kvm, &vm_list, vm_list)
2637                 kvm_for_each_vcpu(i, vcpu, kvm)
2638                         *val += *(u32 *)((void *)vcpu + offset);
2639
2640         raw_spin_unlock(&kvm_lock);
2641         return 0;
2642 }
2643
2644 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2645
2646 static const struct file_operations *stat_fops[] = {
2647         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2648         [KVM_STAT_VM]   = &vm_stat_fops,
2649 };
2650
2651 static void kvm_init_debug(void)
2652 {
2653         struct kvm_stats_debugfs_item *p;
2654
2655         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2656         for (p = debugfs_entries; p->name; ++p)
2657                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2658                                                 (void *)(long)p->offset,
2659                                                 stat_fops[p->kind]);
2660 }
2661
2662 static void kvm_exit_debug(void)
2663 {
2664         struct kvm_stats_debugfs_item *p;
2665
2666         for (p = debugfs_entries; p->name; ++p)
2667                 debugfs_remove(p->dentry);
2668         debugfs_remove(kvm_debugfs_dir);
2669 }
2670
2671 static int kvm_suspend(void)
2672 {
2673         if (kvm_usage_count)
2674                 hardware_disable_nolock(NULL);
2675         return 0;
2676 }
2677
2678 static void kvm_resume(void)
2679 {
2680         if (kvm_usage_count) {
2681                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2682                 hardware_enable_nolock(NULL);
2683         }
2684 }
2685
2686 static struct syscore_ops kvm_syscore_ops = {
2687         .suspend = kvm_suspend,
2688         .resume = kvm_resume,
2689 };
2690
2691 struct page *bad_page;
2692 pfn_t bad_pfn;
2693
2694 static inline
2695 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2696 {
2697         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2698 }
2699
2700 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2701 {
2702         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2703
2704         kvm_arch_vcpu_load(vcpu, cpu);
2705 }
2706
2707 static void kvm_sched_out(struct preempt_notifier *pn,
2708                           struct task_struct *next)
2709 {
2710         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2711
2712         kvm_arch_vcpu_put(vcpu);
2713 }
2714
2715 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2716                   struct module *module)
2717 {
2718         int r;
2719         int cpu;
2720
2721         r = kvm_arch_init(opaque);
2722         if (r)
2723                 goto out_fail;
2724
2725         bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2726
2727         if (bad_page == NULL) {
2728                 r = -ENOMEM;
2729                 goto out;
2730         }
2731
2732         bad_pfn = page_to_pfn(bad_page);
2733
2734         hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2735
2736         if (hwpoison_page == NULL) {
2737                 r = -ENOMEM;
2738                 goto out_free_0;
2739         }
2740
2741         hwpoison_pfn = page_to_pfn(hwpoison_page);
2742
2743         fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2744
2745         if (fault_page == NULL) {
2746                 r = -ENOMEM;
2747                 goto out_free_0;
2748         }
2749
2750         fault_pfn = page_to_pfn(fault_page);
2751
2752         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2753                 r = -ENOMEM;
2754                 goto out_free_0;
2755         }
2756
2757         r = kvm_arch_hardware_setup();
2758         if (r < 0)
2759                 goto out_free_0a;
2760
2761         for_each_online_cpu(cpu) {
2762                 smp_call_function_single(cpu,
2763                                 kvm_arch_check_processor_compat,
2764                                 &r, 1);
2765                 if (r < 0)
2766                         goto out_free_1;
2767         }
2768
2769         r = register_cpu_notifier(&kvm_cpu_notifier);
2770         if (r)
2771                 goto out_free_2;
2772         register_reboot_notifier(&kvm_reboot_notifier);
2773
2774         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2775         if (!vcpu_align)
2776                 vcpu_align = __alignof__(struct kvm_vcpu);
2777         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2778                                            0, NULL);
2779         if (!kvm_vcpu_cache) {
2780                 r = -ENOMEM;
2781                 goto out_free_3;
2782         }
2783
2784         r = kvm_async_pf_init();
2785         if (r)
2786                 goto out_free;
2787
2788         kvm_chardev_ops.owner = module;
2789         kvm_vm_fops.owner = module;
2790         kvm_vcpu_fops.owner = module;
2791
2792         r = misc_register(&kvm_dev);
2793         if (r) {
2794                 printk(KERN_ERR "kvm: misc device register failed\n");
2795                 goto out_unreg;
2796         }
2797
2798         register_syscore_ops(&kvm_syscore_ops);
2799
2800         kvm_preempt_ops.sched_in = kvm_sched_in;
2801         kvm_preempt_ops.sched_out = kvm_sched_out;
2802
2803         kvm_init_debug();
2804
2805         return 0;
2806
2807 out_unreg:
2808         kvm_async_pf_deinit();
2809 out_free:
2810         kmem_cache_destroy(kvm_vcpu_cache);
2811 out_free_3:
2812         unregister_reboot_notifier(&kvm_reboot_notifier);
2813         unregister_cpu_notifier(&kvm_cpu_notifier);
2814 out_free_2:
2815 out_free_1:
2816         kvm_arch_hardware_unsetup();
2817 out_free_0a:
2818         free_cpumask_var(cpus_hardware_enabled);
2819 out_free_0:
2820         if (fault_page)
2821                 __free_page(fault_page);
2822         if (hwpoison_page)
2823                 __free_page(hwpoison_page);
2824         __free_page(bad_page);
2825 out:
2826         kvm_arch_exit();
2827 out_fail:
2828         return r;
2829 }
2830 EXPORT_SYMBOL_GPL(kvm_init);
2831
2832 void kvm_exit(void)
2833 {
2834         kvm_exit_debug();
2835         misc_deregister(&kvm_dev);
2836         kmem_cache_destroy(kvm_vcpu_cache);
2837         kvm_async_pf_deinit();
2838         unregister_syscore_ops(&kvm_syscore_ops);
2839         unregister_reboot_notifier(&kvm_reboot_notifier);
2840         unregister_cpu_notifier(&kvm_cpu_notifier);
2841         on_each_cpu(hardware_disable_nolock, NULL, 1);
2842         kvm_arch_hardware_unsetup();
2843         kvm_arch_exit();
2844         free_cpumask_var(cpus_hardware_enabled);
2845         __free_page(hwpoison_page);
2846         __free_page(bad_page);
2847 }
2848 EXPORT_SYMBOL_GPL(kvm_exit);