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 bool kvm_is_mmio_pfn(pfn_t pfn)
 104 {
 105         if (pfn_valid(pfn)) {
 106                 int reserved;
 107                 struct page *tail = pfn_to_page(pfn);
 108                 struct page *head = compound_trans_head(tail);
 109                 reserved = PageReserved(head);
 110                 if (head != tail) {
 111                         /*
 112                          * "head" is not a dangling pointer
 113                          * (compound_trans_head takes care of that)
 114                          * but the hugepage may have been splitted
 115                          * from under us (and we may not hold a
 116                          * reference count on the head page so it can
 117                          * be reused before we run PageReferenced), so
 118                          * we've to check PageTail before returning
 119                          * what we just read.
 120                          */
 121                         smp_rmb();
 122                         if (PageTail(tail))
 123                                 return reserved;
 124                 }
 125                 return PageReserved(tail);
 126         }
 127
 128         return true;
 129 }
 130
 131 /*
 132  * Switches to specified vcpu, until a matching vcpu_put()
 133  */
 134 void vcpu_load(struct kvm_vcpu *vcpu)
 135 {
 136         int cpu;
 137
 138         mutex_lock(&vcpu->mutex);
 139         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 140                 /* The thread running this VCPU changed. */
 141                 struct pid *oldpid = vcpu->pid;
 142                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 143                 rcu_assign_pointer(vcpu->pid, newpid);
 144                 synchronize_rcu();
 145                 put_pid(oldpid);
 146         }
 147         cpu = get_cpu();
 148         preempt_notifier_register(&vcpu->preempt_notifier);
 149         kvm_arch_vcpu_load(vcpu, cpu);
 150         put_cpu();
 151 }
 152
 153 void vcpu_put(struct kvm_vcpu *vcpu)
 154 {
 155         preempt_disable();
 156         kvm_arch_vcpu_put(vcpu);
 157         preempt_notifier_unregister(&vcpu->preempt_notifier);
 158         preempt_enable();
 159         mutex_unlock(&vcpu->mutex);
 160 }
 161
 162 static void ack_flush(void *_completed)
 163 {
 164 }
 165
 166 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 167 {
 168         int i, cpu, me;
 169         cpumask_var_t cpus;
 170         bool called = true;
 171         struct kvm_vcpu *vcpu;
 172
 173         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 174
 175         me = get_cpu();
 176         kvm_for_each_vcpu(i, vcpu, kvm) {
 177                 kvm_make_request(req, vcpu);
 178                 cpu = vcpu->cpu;
 179
 180                 /* Set ->requests bit before we read ->mode */
 181                 smp_mb();
 182
 183                 if (cpus != NULL && cpu != -1 && cpu != me &&
 184                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 185                         cpumask_set_cpu(cpu, cpus);
 186         }
 187         if (unlikely(cpus == NULL))
 188                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 189         else if (!cpumask_empty(cpus))
 190                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 191         else
 192                 called = false;
 193         put_cpu();
 194         free_cpumask_var(cpus);
 195         return called;
 196 }
 197
 198 void kvm_flush_remote_tlbs(struct kvm *kvm)
 199 {
 200         long dirty_count = kvm->tlbs_dirty;
 201
 202         smp_mb();
 203         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 204                 ++kvm->stat.remote_tlb_flush;
 205         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 206 }
 207
 208 void kvm_reload_remote_mmus(struct kvm *kvm)
 209 {
 210         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 211 }
 212
 213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 214 {
 215         struct page *page;
 216         int r;
 217
 218         mutex_init(&vcpu->mutex);
 219         vcpu->cpu = -1;
 220         vcpu->kvm = kvm;
 221         vcpu->vcpu_id = id;
 222         vcpu->pid = NULL;
 223         init_waitqueue_head(&vcpu->wq);
 224         kvm_async_pf_vcpu_init(vcpu);
 225
 226         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 227         if (!page) {
 228                 r = -ENOMEM;
 229                 goto fail;
 230         }
 231         vcpu->run = page_address(page);
 232
 233         kvm_vcpu_set_in_spin_loop(vcpu, false);
 234         kvm_vcpu_set_dy_eligible(vcpu, false);
 235
 236         r = kvm_arch_vcpu_init(vcpu);
 237         if (r < 0)
 238                 goto fail_free_run;
 239         return 0;
 240
 241 fail_free_run:
 242         free_page((unsigned long)vcpu->run);
 243 fail:
 244         return r;
 245 }
 246 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 247
 248 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 249 {
 250         put_pid(vcpu->pid);
 251         kvm_arch_vcpu_uninit(vcpu);
 252         free_page((unsigned long)vcpu->run);
 253 }
 254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 255
 256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 257 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 258 {
 259         return container_of(mn, struct kvm, mmu_notifier);
 260 }
 261
 262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 263                                              struct mm_struct *mm,
 264                                              unsigned long address)
 265 {
 266         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 267         int need_tlb_flush, idx;
 268
 269         /*
 270          * When ->invalidate_page runs, the linux pte has been zapped
 271          * already but the page is still allocated until
 272          * ->invalidate_page returns. So if we increase the sequence
 273          * here the kvm page fault will notice if the spte can't be
 274          * established because the page is going to be freed. If
 275          * instead the kvm page fault establishes the spte before
 276          * ->invalidate_page runs, kvm_unmap_hva will release it
 277          * before returning.
 278          *
 279          * The sequence increase only need to be seen at spin_unlock
 280          * time, and not at spin_lock time.
 281          *
 282          * Increasing the sequence after the spin_unlock would be
 283          * unsafe because the kvm page fault could then establish the
 284          * pte after kvm_unmap_hva returned, without noticing the page
 285          * is going to be freed.
 286          */
 287         idx = srcu_read_lock(&kvm->srcu);
 288         spin_lock(&kvm->mmu_lock);
 289
 290         kvm->mmu_notifier_seq++;
 291         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 292         /* we've to flush the tlb before the pages can be freed */
 293         if (need_tlb_flush)
 294                 kvm_flush_remote_tlbs(kvm);
 295
 296         spin_unlock(&kvm->mmu_lock);
 297         srcu_read_unlock(&kvm->srcu, idx);
 298 }
 299
 300 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 301                                         struct mm_struct *mm,
 302                                         unsigned long address,
 303                                         pte_t pte)
 304 {
 305         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 306         int idx;
 307
 308         idx = srcu_read_lock(&kvm->srcu);
 309         spin_lock(&kvm->mmu_lock);
 310         kvm->mmu_notifier_seq++;
 311         kvm_set_spte_hva(kvm, address, pte);
 312         spin_unlock(&kvm->mmu_lock);
 313         srcu_read_unlock(&kvm->srcu, idx);
 314 }
 315
 316 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 317                                                     struct mm_struct *mm,
 318                                                     unsigned long start,
 319                                                     unsigned long end)
 320 {
 321         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 322         int need_tlb_flush = 0, idx;
 323
 324         idx = srcu_read_lock(&kvm->srcu);
 325         spin_lock(&kvm->mmu_lock);
 326         /*
 327          * The count increase must become visible at unlock time as no
 328          * spte can be established without taking the mmu_lock and
 329          * count is also read inside the mmu_lock critical section.
 330          */
 331         kvm->mmu_notifier_count++;
 332         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 333         need_tlb_flush |= kvm->tlbs_dirty;
 334         /* we've to flush the tlb before the pages can be freed */
 335         if (need_tlb_flush)
 336                 kvm_flush_remote_tlbs(kvm);
 337
 338         spin_unlock(&kvm->mmu_lock);
 339         srcu_read_unlock(&kvm->srcu, idx);
 340 }
 341
 342 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 343                                                   struct mm_struct *mm,
 344                                                   unsigned long start,
 345                                                   unsigned long end)
 346 {
 347         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 348
 349         spin_lock(&kvm->mmu_lock);
 350         /*
 351          * This sequence increase will notify the kvm page fault that
 352          * the page that is going to be mapped in the spte could have
 353          * been freed.
 354          */
 355         kvm->mmu_notifier_seq++;
 356         smp_wmb();
 357         /*
 358          * The above sequence increase must be visible before the
 359          * below count decrease, which is ensured by the smp_wmb above
 360          * in conjunction with the smp_rmb in mmu_notifier_retry().
 361          */
 362         kvm->mmu_notifier_count--;
 363         spin_unlock(&kvm->mmu_lock);
 364
 365         BUG_ON(kvm->mmu_notifier_count < 0);
 366 }
 367
 368 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 369                                               struct mm_struct *mm,
 370                                               unsigned long address)
 371 {
 372         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 373         int young, idx;
 374
 375         idx = srcu_read_lock(&kvm->srcu);
 376         spin_lock(&kvm->mmu_lock);
 377
 378         young = kvm_age_hva(kvm, address);
 379         if (young)
 380                 kvm_flush_remote_tlbs(kvm);
 381
 382         spin_unlock(&kvm->mmu_lock);
 383         srcu_read_unlock(&kvm->srcu, idx);
 384
 385         return young;
 386 }
 387
 388 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 389                                        struct mm_struct *mm,
 390                                        unsigned long address)
 391 {
 392         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 393         int young, idx;
 394
 395         idx = srcu_read_lock(&kvm->srcu);
 396         spin_lock(&kvm->mmu_lock);
 397         young = kvm_test_age_hva(kvm, address);
 398         spin_unlock(&kvm->mmu_lock);
 399         srcu_read_unlock(&kvm->srcu, idx);
 400
 401         return young;
 402 }
 403
 404 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 405                                      struct mm_struct *mm)
 406 {
 407         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 408         int idx;
 409
 410         idx = srcu_read_lock(&kvm->srcu);
 411         kvm_arch_flush_shadow(kvm);
 412         srcu_read_unlock(&kvm->srcu, idx);
 413 }
 414
 415 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 416         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 417         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 418         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 419         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 420         .test_young             = kvm_mmu_notifier_test_young,
 421         .change_pte             = kvm_mmu_notifier_change_pte,
 422         .release                = kvm_mmu_notifier_release,
 423 };
 424
 425 static int kvm_init_mmu_notifier(struct kvm *kvm)
 426 {
 427         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 428         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 429 }
 430
 431 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 432
 433 static int kvm_init_mmu_notifier(struct kvm *kvm)
 434 {
 435         return 0;
 436 }
 437
 438 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 439
 440 static void kvm_init_memslots_id(struct kvm *kvm)
 441 {
 442         int i;
 443         struct kvm_memslots *slots = kvm->memslots;
 444
 445         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 446                 slots->id_to_index[i] = slots->memslots[i].id = i;
 447 }
 448
 449 static struct kvm *kvm_create_vm(unsigned long type)
 450 {
 451         int r, i;
 452         struct kvm *kvm = kvm_arch_alloc_vm();
 453
 454         if (!kvm)
 455                 return ERR_PTR(-ENOMEM);
 456
 457         r = kvm_arch_init_vm(kvm, type);
 458         if (r)
 459                 goto out_err_nodisable;
 460
 461         r = hardware_enable_all();
 462         if (r)
 463                 goto out_err_nodisable;
 464
 465 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 466         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 467         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 468 #endif
 469
 470         r = -ENOMEM;
 471         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 472         if (!kvm->memslots)
 473                 goto out_err_nosrcu;
 474         kvm_init_memslots_id(kvm);
 475         if (init_srcu_struct(&kvm->srcu))
 476                 goto out_err_nosrcu;
 477         for (i = 0; i < KVM_NR_BUSES; i++) {
 478                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 479                                         GFP_KERNEL);
 480                 if (!kvm->buses[i])
 481                         goto out_err;
 482         }
 483
 484         spin_lock_init(&kvm->mmu_lock);
 485         kvm->mm = current->mm;
 486         atomic_inc(&kvm->mm->mm_count);
 487         kvm_eventfd_init(kvm);
 488         mutex_init(&kvm->lock);
 489         mutex_init(&kvm->irq_lock);
 490         mutex_init(&kvm->slots_lock);
 491         atomic_set(&kvm->users_count, 1);
 492
 493         r = kvm_init_mmu_notifier(kvm);
 494         if (r)
 495                 goto out_err;
 496
 497         raw_spin_lock(&kvm_lock);
 498         list_add(&kvm->vm_list, &vm_list);
 499         raw_spin_unlock(&kvm_lock);
 500
 501         return kvm;
 502
 503 out_err:
 504         cleanup_srcu_struct(&kvm->srcu);
 505 out_err_nosrcu:
 506         hardware_disable_all();
 507 out_err_nodisable:
 508         for (i = 0; i < KVM_NR_BUSES; i++)
 509                 kfree(kvm->buses[i]);
 510         kfree(kvm->memslots);
 511         kvm_arch_free_vm(kvm);
 512         return ERR_PTR(r);
 513 }
 514
 515 /*
 516  * Avoid using vmalloc for a small buffer.
 517  * Should not be used when the size is statically known.
 518  */
 519 void *kvm_kvzalloc(unsigned long size)
 520 {
 521         if (size > PAGE_SIZE)
 522                 return vzalloc(size);
 523         else
 524                 return kzalloc(size, GFP_KERNEL);
 525 }
 526
 527 void kvm_kvfree(const void *addr)
 528 {
 529         if (is_vmalloc_addr(addr))
 530                 vfree(addr);
 531         else
 532                 kfree(addr);
 533 }
 534
 535 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 536 {
 537         if (!memslot->dirty_bitmap)
 538                 return;
 539
 540         kvm_kvfree(memslot->dirty_bitmap);
 541         memslot->dirty_bitmap = NULL;
 542 }
 543
 544 /*
 545  * Free any memory in @free but not in @dont.
 546  */
 547 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 548                                   struct kvm_memory_slot *dont)
 549 {
 550         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 551                 kvm_destroy_dirty_bitmap(free);
 552
 553         kvm_arch_free_memslot(free, dont);
 554
 555         free->npages = 0;
 556 }
 557
 558 void kvm_free_physmem(struct kvm *kvm)
 559 {
 560         struct kvm_memslots *slots = kvm->memslots;
 561         struct kvm_memory_slot *memslot;
 562
 563         kvm_for_each_memslot(memslot, slots)
 564                 kvm_free_physmem_slot(memslot, NULL);
 565
 566         kfree(kvm->memslots);
 567 }
 568
 569 static void kvm_destroy_vm(struct kvm *kvm)
 570 {
 571         int i;
 572         struct mm_struct *mm = kvm->mm;
 573
 574         kvm_arch_sync_events(kvm);
 575         raw_spin_lock(&kvm_lock);
 576         list_del(&kvm->vm_list);
 577         raw_spin_unlock(&kvm_lock);
 578         kvm_free_irq_routing(kvm);
 579         for (i = 0; i < KVM_NR_BUSES; i++)
 580                 kvm_io_bus_destroy(kvm->buses[i]);
 581         kvm_coalesced_mmio_free(kvm);
 582 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 583         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 584 #else
 585         kvm_arch_flush_shadow(kvm);
 586 #endif
 587         kvm_arch_destroy_vm(kvm);
 588         kvm_free_physmem(kvm);
 589         cleanup_srcu_struct(&kvm->srcu);
 590         kvm_arch_free_vm(kvm);
 591         hardware_disable_all();
 592         mmdrop(mm);
 593 }
 594
 595 void kvm_get_kvm(struct kvm *kvm)
 596 {
 597         atomic_inc(&kvm->users_count);
 598 }
 599 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 600
 601 void kvm_put_kvm(struct kvm *kvm)
 602 {
 603         if (atomic_dec_and_test(&kvm->users_count))
 604                 kvm_destroy_vm(kvm);
 605 }
 606 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 607
 608
 609 static int kvm_vm_release(struct inode *inode, struct file *filp)
 610 {
 611         struct kvm *kvm = filp->private_data;
 612
 613         kvm_irqfd_release(kvm);
 614
 615         kvm_put_kvm(kvm);
 616         return 0;
 617 }
 618
 619 /*
 620  * Allocation size is twice as large as the actual dirty bitmap size.
 621  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 622  */
 623 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 624 {
 625 #ifndef CONFIG_S390
 626         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 627
 628         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 629         if (!memslot->dirty_bitmap)
 630                 return -ENOMEM;
 631
 632 #endif /* !CONFIG_S390 */
 633         return 0;
 634 }
 635
 636 static int cmp_memslot(const void *slot1, const void *slot2)
 637 {
 638         struct kvm_memory_slot *s1, *s2;
 639
 640         s1 = (struct kvm_memory_slot *)slot1;
 641         s2 = (struct kvm_memory_slot *)slot2;
 642
 643         if (s1->npages < s2->npages)
 644                 return 1;
 645         if (s1->npages > s2->npages)
 646                 return -1;
 647
 648         return 0;
 649 }
 650
 651 /*
 652  * Sort the memslots base on its size, so the larger slots
 653  * will get better fit.
 654  */
 655 static void sort_memslots(struct kvm_memslots *slots)
 656 {
 657         int i;
 658
 659         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 660               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 661
 662         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 663                 slots->id_to_index[slots->memslots[i].id] = i;
 664 }
 665
 666 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 667 {
 668         if (new) {
 669                 int id = new->id;
 670                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 671                 unsigned long npages = old->npages;
 672
 673                 *old = *new;
 674                 if (new->npages != npages)
 675                         sort_memslots(slots);
 676         }
 677
 678         slots->generation++;
 679 }
 680
 681 static int check_memory_region_flags(struct kvm_userspace_memory_region *mem)
 682 {
 683         if (mem->flags & ~KVM_MEM_LOG_DIRTY_PAGES)
 684                 return -EINVAL;
 685
 686         return 0;
 687 }
 688
 689 /*
 690  * Allocate some memory and give it an address in the guest physical address
 691  * space.
 692  *
 693  * Discontiguous memory is allowed, mostly for framebuffers.
 694  *
 695  * Must be called holding mmap_sem for write.
 696  */
 697 int __kvm_set_memory_region(struct kvm *kvm,
 698                             struct kvm_userspace_memory_region *mem,
 699                             int user_alloc)
 700 {
 701         int r;
 702         gfn_t base_gfn;
 703         unsigned long npages;
 704         unsigned long i;
 705         struct kvm_memory_slot *memslot;
 706         struct kvm_memory_slot old, new;
 707         struct kvm_memslots *slots, *old_memslots;
 708
 709         r = check_memory_region_flags(mem);
 710         if (r)
 711                 goto out;
 712
 713         r = -EINVAL;
 714         /* General sanity checks */
 715         if (mem->memory_size & (PAGE_SIZE - 1))
 716                 goto out;
 717         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 718                 goto out;
 719         /* We can read the guest memory with __xxx_user() later on. */
 720         if (user_alloc &&
 721             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 722              !access_ok(VERIFY_WRITE,
 723                         (void __user *)(unsigned long)mem->userspace_addr,
 724                         mem->memory_size)))
 725                 goto out;
 726         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 727                 goto out;
 728         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 729                 goto out;
 730
 731         memslot = id_to_memslot(kvm->memslots, mem->slot);
 732         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 733         npages = mem->memory_size >> PAGE_SHIFT;
 734
 735         r = -EINVAL;
 736         if (npages > KVM_MEM_MAX_NR_PAGES)
 737                 goto out;
 738
 739         if (!npages)
 740                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 741
 742         new = old = *memslot;
 743
 744         new.id = mem->slot;
 745         new.base_gfn = base_gfn;
 746         new.npages = npages;
 747         new.flags = mem->flags;
 748
 749         /* Disallow changing a memory slot's size. */
 750         r = -EINVAL;
 751         if (npages && old.npages && npages != old.npages)
 752                 goto out_free;
 753
 754         /* Check for overlaps */
 755         r = -EEXIST;
 756         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 757                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 758
 759                 if (s == memslot || !s->npages)
 760                         continue;
 761                 if (!((base_gfn + npages <= s->base_gfn) ||
 762                       (base_gfn >= s->base_gfn + s->npages)))
 763                         goto out_free;
 764         }
 765
 766         /* Free page dirty bitmap if unneeded */
 767         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 768                 new.dirty_bitmap = NULL;
 769
 770         r = -ENOMEM;
 771
 772         /* Allocate if a slot is being created */
 773         if (npages && !old.npages) {
 774                 new.user_alloc = user_alloc;
 775                 new.userspace_addr = mem->userspace_addr;
 776
 777                 if (kvm_arch_create_memslot(&new, npages))
 778                         goto out_free;
 779         }
 780
 781         /* Allocate page dirty bitmap if needed */
 782         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 783                 if (kvm_create_dirty_bitmap(&new) < 0)
 784                         goto out_free;
 785                 /* destroy any largepage mappings for dirty tracking */
 786         }
 787
 788         if (!npages) {
 789                 struct kvm_memory_slot *slot;
 790
 791                 r = -ENOMEM;
 792                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 793                                 GFP_KERNEL);
 794                 if (!slots)
 795                         goto out_free;
 796                 slot = id_to_memslot(slots, mem->slot);
 797                 slot->flags |= KVM_MEMSLOT_INVALID;
 798
 799                 update_memslots(slots, NULL);
 800
 801                 old_memslots = kvm->memslots;
 802                 rcu_assign_pointer(kvm->memslots, slots);
 803                 synchronize_srcu_expedited(&kvm->srcu);
 804                 /* From this point no new shadow pages pointing to a deleted
 805                  * memslot will be created.
 806                  *
 807                  * validation of sp->gfn happens in:
 808                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 809                  *      - kvm_is_visible_gfn (mmu_check_roots)
 810                  */
 811                 kvm_arch_flush_shadow(kvm);
 812                 kfree(old_memslots);
 813         }
 814
 815         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 816         if (r)
 817                 goto out_free;
 818
 819         /* map/unmap the pages in iommu page table */
 820         if (npages) {
 821                 r = kvm_iommu_map_pages(kvm, &new);
 822                 if (r)
 823                         goto out_free;
 824         } else
 825                 kvm_iommu_unmap_pages(kvm, &old);
 826
 827         r = -ENOMEM;
 828         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 829                         GFP_KERNEL);
 830         if (!slots)
 831                 goto out_free;
 832
 833         /* actual memory is freed via old in kvm_free_physmem_slot below */
 834         if (!npages) {
 835                 new.dirty_bitmap = NULL;
 836                 memset(&new.arch, 0, sizeof(new.arch));
 837         }
 838
 839         update_memslots(slots, &new);
 840         old_memslots = kvm->memslots;
 841         rcu_assign_pointer(kvm->memslots, slots);
 842         synchronize_srcu_expedited(&kvm->srcu);
 843
 844         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 845
 846         /*
 847          * If the new memory slot is created, we need to clear all
 848          * mmio sptes.
 849          */
 850         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 851                 kvm_arch_flush_shadow(kvm);
 852
 853         kvm_free_physmem_slot(&old, &new);
 854         kfree(old_memslots);
 855
 856         return 0;
 857
 858 out_free:
 859         kvm_free_physmem_slot(&new, &old);
 860 out:
 861         return r;
 862
 863 }
 864 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 865
 866 int kvm_set_memory_region(struct kvm *kvm,
 867                           struct kvm_userspace_memory_region *mem,
 868                           int user_alloc)
 869 {
 870         int r;
 871
 872         mutex_lock(&kvm->slots_lock);
 873         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 874         mutex_unlock(&kvm->slots_lock);
 875         return r;
 876 }
 877 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 878
 879 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 880                                    struct
 881                                    kvm_userspace_memory_region *mem,
 882                                    int user_alloc)
 883 {
 884         if (mem->slot >= KVM_MEMORY_SLOTS)
 885                 return -EINVAL;
 886         return kvm_set_memory_region(kvm, mem, user_alloc);
 887 }
 888
 889 int kvm_get_dirty_log(struct kvm *kvm,
 890                         struct kvm_dirty_log *log, int *is_dirty)
 891 {
 892         struct kvm_memory_slot *memslot;
 893         int r, i;
 894         unsigned long n;
 895         unsigned long any = 0;
 896
 897         r = -EINVAL;
 898         if (log->slot >= KVM_MEMORY_SLOTS)
 899                 goto out;
 900
 901         memslot = id_to_memslot(kvm->memslots, log->slot);
 902         r = -ENOENT;
 903         if (!memslot->dirty_bitmap)
 904                 goto out;
 905
 906         n = kvm_dirty_bitmap_bytes(memslot);
 907
 908         for (i = 0; !any && i < n/sizeof(long); ++i)
 909                 any = memslot->dirty_bitmap[i];
 910
 911         r = -EFAULT;
 912         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 913                 goto out;
 914
 915         if (any)
 916                 *is_dirty = 1;
 917
 918         r = 0;
 919 out:
 920         return r;
 921 }
 922
 923 bool kvm_largepages_enabled(void)
 924 {
 925         return largepages_enabled;
 926 }
 927
 928 void kvm_disable_largepages(void)
 929 {
 930         largepages_enabled = false;
 931 }
 932 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 933
 934 static inline unsigned long bad_hva(void)
 935 {
 936         return PAGE_OFFSET;
 937 }
 938
 939 int kvm_is_error_hva(unsigned long addr)
 940 {
 941         return addr == bad_hva();
 942 }
 943 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 944
 945 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 946 {
 947         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 948 }
 949 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 950
 951 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 952 {
 953         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 954
 955         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 956               memslot->flags & KVM_MEMSLOT_INVALID)
 957                 return 0;
 958
 959         return 1;
 960 }
 961 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 962
 963 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 964 {
 965         struct vm_area_struct *vma;
 966         unsigned long addr, size;
 967
 968         size = PAGE_SIZE;
 969
 970         addr = gfn_to_hva(kvm, gfn);
 971         if (kvm_is_error_hva(addr))
 972                 return PAGE_SIZE;
 973
 974         down_read(&current->mm->mmap_sem);
 975         vma = find_vma(current->mm, addr);
 976         if (!vma)
 977                 goto out;
 978
 979         size = vma_kernel_pagesize(vma);
 980
 981 out:
 982         up_read(&current->mm->mmap_sem);
 983
 984         return size;
 985 }
 986
 987 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
 988                                      gfn_t *nr_pages)
 989 {
 990         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 991                 return bad_hva();
 992
 993         if (nr_pages)
 994                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
 995
 996         return gfn_to_hva_memslot(slot, gfn);
 997 }
 998
 999 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1000 {
1001         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1002 }
1003 EXPORT_SYMBOL_GPL(gfn_to_hva);
1004
1005 /*
1006  * The hva returned by this function is only allowed to be read.
1007  * It should pair with kvm_read_hva() or kvm_read_hva_atomic().
1008  */
1009 static unsigned long gfn_to_hva_read(struct kvm *kvm, gfn_t gfn)
1010 {
1011         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1012 }
1013
1014 static int kvm_read_hva(void *data, void __user *hva, int len)
1015 {
1016         return __copy_from_user(data, hva, len);
1017 }
1018
1019 static int kvm_read_hva_atomic(void *data, void __user *hva, int len)
1020 {
1021         return __copy_from_user_inatomic(data, hva, len);
1022 }
1023
1024 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1025         unsigned long start, int write, struct page **page)
1026 {
1027         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1028
1029         if (write)
1030                 flags |= FOLL_WRITE;
1031
1032         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1033 }
1034
1035 static inline int check_user_page_hwpoison(unsigned long addr)
1036 {
1037         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1038
1039         rc = __get_user_pages(current, current->mm, addr, 1,
1040                               flags, NULL, NULL, NULL);
1041         return rc == -EHWPOISON;
1042 }
1043
1044 /*
1045  * The atomic path to get the writable pfn which will be stored in @pfn,
1046  * true indicates success, otherwise false is returned.
1047  */
1048 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1049                             bool write_fault, bool *writable, pfn_t *pfn)
1050 {
1051         struct page *page[1];
1052         int npages;
1053
1054         if (!(async || atomic))
1055                 return false;
1056
1057         /*
1058          * Fast pin a writable pfn only if it is a write fault request
1059          * or the caller allows to map a writable pfn for a read fault
1060          * request.
1061          */
1062         if (!(write_fault || writable))
1063                 return false;
1064
1065         npages = __get_user_pages_fast(addr, 1, 1, page);
1066         if (npages == 1) {
1067                 *pfn = page_to_pfn(page[0]);
1068
1069                 if (writable)
1070                         *writable = true;
1071                 return true;
1072         }
1073
1074         return false;
1075 }
1076
1077 /*
1078  * The slow path to get the pfn of the specified host virtual address,
1079  * 1 indicates success, -errno is returned if error is detected.
1080  */
1081 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1082                            bool *writable, pfn_t *pfn)
1083 {
1084         struct page *page[1];
1085         int npages = 0;
1086
1087         might_sleep();
1088
1089         if (writable)
1090                 *writable = write_fault;
1091
1092         if (async) {
1093                 down_read(&current->mm->mmap_sem);
1094                 npages = get_user_page_nowait(current, current->mm,
1095                                               addr, write_fault, page);
1096                 up_read(&current->mm->mmap_sem);
1097         } else
1098                 npages = get_user_pages_fast(addr, 1, write_fault,
1099                                              page);
1100         if (npages != 1)
1101                 return npages;
1102
1103         /* map read fault as writable if possible */
1104         if (unlikely(!write_fault) && writable) {
1105                 struct page *wpage[1];
1106
1107                 npages = __get_user_pages_fast(addr, 1, 1, wpage);
1108                 if (npages == 1) {
1109                         *writable = true;
1110                         put_page(page[0]);
1111                         page[0] = wpage[0];
1112                 }
1113
1114                 npages = 1;
1115         }
1116         *pfn = page_to_pfn(page[0]);
1117         return npages;
1118 }
1119
1120 /*
1121  * Pin guest page in memory and return its pfn.
1122  * @addr: host virtual address which maps memory to the guest
1123  * @atomic: whether this function can sleep
1124  * @async: whether this function need to wait IO complete if the
1125  *         host page is not in the memory
1126  * @write_fault: whether we should get a writable host page
1127  * @writable: whether it allows to map a writable host page for !@write_fault
1128  *
1129  * The function will map a writable host page for these two cases:
1130  * 1): @write_fault = true
1131  * 2): @write_fault = false && @writable, @writable will tell the caller
1132  *     whether the mapping is writable.
1133  */
1134 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1135                         bool write_fault, bool *writable)
1136 {
1137         struct vm_area_struct *vma;
1138         pfn_t pfn = 0;
1139         int npages;
1140
1141         /* we can do it either atomically or asynchronously, not both */
1142         BUG_ON(atomic && async);
1143
1144         BUG_ON(!write_fault && !writable);
1145
1146         if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1147                 return pfn;
1148
1149         if (atomic)
1150                 return KVM_PFN_ERR_FAULT;
1151
1152         npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1153         if (npages == 1)
1154                 return pfn;
1155
1156         down_read(&current->mm->mmap_sem);
1157         if (npages == -EHWPOISON ||
1158               (!async && check_user_page_hwpoison(addr))) {
1159                 pfn = KVM_PFN_ERR_HWPOISON;
1160                 goto exit;
1161         }
1162
1163         vma = find_vma_intersection(current->mm, addr, addr + 1);
1164
1165         if (vma == NULL)
1166                 pfn = KVM_PFN_ERR_FAULT;
1167         else if ((vma->vm_flags & VM_PFNMAP)) {
1168                 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1169                         vma->vm_pgoff;
1170                 BUG_ON(!kvm_is_mmio_pfn(pfn));
1171         } else {
1172                 if (async && (vma->vm_flags & VM_WRITE))
1173                         *async = true;
1174                 pfn = KVM_PFN_ERR_FAULT;
1175         }
1176 exit:
1177         up_read(&current->mm->mmap_sem);
1178         return pfn;
1179 }
1180
1181 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1182                           bool write_fault, bool *writable)
1183 {
1184         unsigned long addr;
1185
1186         if (async)
1187                 *async = false;
1188
1189         addr = gfn_to_hva(kvm, gfn);
1190         if (kvm_is_error_hva(addr))
1191                 return KVM_PFN_ERR_BAD;
1192
1193         return hva_to_pfn(addr, atomic, async, write_fault, writable);
1194 }
1195
1196 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1197 {
1198         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1199 }
1200 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1201
1202 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1203                        bool write_fault, bool *writable)
1204 {
1205         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1206 }
1207 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1208
1209 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1210 {
1211         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1212 }
1213 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1214
1215 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1216                       bool *writable)
1217 {
1218         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1219 }
1220 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1221
1222 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1223 {
1224         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1225         return hva_to_pfn(addr, false, NULL, true, NULL);
1226 }
1227
1228 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1229 {
1230         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1231
1232         return hva_to_pfn(addr, true, NULL, true, NULL);
1233 }
1234 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1235
1236 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1237                                                                   int nr_pages)
1238 {
1239         unsigned long addr;
1240         gfn_t entry;
1241
1242         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1243         if (kvm_is_error_hva(addr))
1244                 return -1;
1245
1246         if (entry < nr_pages)
1247                 return 0;
1248
1249         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1250 }
1251 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1252
1253 static struct page *kvm_pfn_to_page(pfn_t pfn)
1254 {
1255         if (is_error_pfn(pfn))
1256                 return KVM_ERR_PTR_BAD_PAGE;
1257
1258         if (kvm_is_mmio_pfn(pfn)) {
1259                 WARN_ON(1);
1260                 return KVM_ERR_PTR_BAD_PAGE;
1261         }
1262
1263         return pfn_to_page(pfn);
1264 }
1265
1266 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1267 {
1268         pfn_t pfn;
1269
1270         pfn = gfn_to_pfn(kvm, gfn);
1271
1272         return kvm_pfn_to_page(pfn);
1273 }
1274
1275 EXPORT_SYMBOL_GPL(gfn_to_page);
1276
1277 void kvm_release_page_clean(struct page *page)
1278 {
1279         WARN_ON(is_error_page(page));
1280
1281         kvm_release_pfn_clean(page_to_pfn(page));
1282 }
1283 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1284
1285 void kvm_release_pfn_clean(pfn_t pfn)
1286 {
1287         WARN_ON(is_error_pfn(pfn));
1288
1289         if (!kvm_is_mmio_pfn(pfn))
1290                 put_page(pfn_to_page(pfn));
1291 }
1292 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1293
1294 void kvm_release_page_dirty(struct page *page)
1295 {
1296         WARN_ON(is_error_page(page));
1297
1298         kvm_release_pfn_dirty(page_to_pfn(page));
1299 }
1300 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1301
1302 void kvm_release_pfn_dirty(pfn_t pfn)
1303 {
1304         kvm_set_pfn_dirty(pfn);
1305         kvm_release_pfn_clean(pfn);
1306 }
1307 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1308
1309 void kvm_set_page_dirty(struct page *page)
1310 {
1311         kvm_set_pfn_dirty(page_to_pfn(page));
1312 }
1313 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1314
1315 void kvm_set_pfn_dirty(pfn_t pfn)
1316 {
1317         if (!kvm_is_mmio_pfn(pfn)) {
1318                 struct page *page = pfn_to_page(pfn);
1319                 if (!PageReserved(page))
1320                         SetPageDirty(page);
1321         }
1322 }
1323 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1324
1325 void kvm_set_pfn_accessed(pfn_t pfn)
1326 {
1327         if (!kvm_is_mmio_pfn(pfn))
1328                 mark_page_accessed(pfn_to_page(pfn));
1329 }
1330 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1331
1332 void kvm_get_pfn(pfn_t pfn)
1333 {
1334         if (!kvm_is_mmio_pfn(pfn))
1335                 get_page(pfn_to_page(pfn));
1336 }
1337 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1338
1339 static int next_segment(unsigned long len, int offset)
1340 {
1341         if (len > PAGE_SIZE - offset)
1342                 return PAGE_SIZE - offset;
1343         else
1344                 return len;
1345 }
1346
1347 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1348                         int len)
1349 {
1350         int r;
1351         unsigned long addr;
1352
1353         addr = gfn_to_hva_read(kvm, gfn);
1354         if (kvm_is_error_hva(addr))
1355                 return -EFAULT;
1356         r = kvm_read_hva(data, (void __user *)addr + offset, len);
1357         if (r)
1358                 return -EFAULT;
1359         return 0;
1360 }
1361 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1362
1363 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1364 {
1365         gfn_t gfn = gpa >> PAGE_SHIFT;
1366         int seg;
1367         int offset = offset_in_page(gpa);
1368         int ret;
1369
1370         while ((seg = next_segment(len, offset)) != 0) {
1371                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1372                 if (ret < 0)
1373                         return ret;
1374                 offset = 0;
1375                 len -= seg;
1376                 data += seg;
1377                 ++gfn;
1378         }
1379         return 0;
1380 }
1381 EXPORT_SYMBOL_GPL(kvm_read_guest);
1382
1383 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1384                           unsigned long len)
1385 {
1386         int r;
1387         unsigned long addr;
1388         gfn_t gfn = gpa >> PAGE_SHIFT;
1389         int offset = offset_in_page(gpa);
1390
1391         addr = gfn_to_hva_read(kvm, gfn);
1392         if (kvm_is_error_hva(addr))
1393                 return -EFAULT;
1394         pagefault_disable();
1395         r = kvm_read_hva_atomic(data, (void __user *)addr + offset, len);
1396         pagefault_enable();
1397         if (r)
1398                 return -EFAULT;
1399         return 0;
1400 }
1401 EXPORT_SYMBOL(kvm_read_guest_atomic);
1402
1403 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1404                          int offset, int len)
1405 {
1406         int r;
1407         unsigned long addr;
1408
1409         addr = gfn_to_hva(kvm, gfn);
1410         if (kvm_is_error_hva(addr))
1411                 return -EFAULT;
1412         r = __copy_to_user((void __user *)addr + offset, data, len);
1413         if (r)
1414                 return -EFAULT;
1415         mark_page_dirty(kvm, gfn);
1416         return 0;
1417 }
1418 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1419
1420 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1421                     unsigned long len)
1422 {
1423         gfn_t gfn = gpa >> PAGE_SHIFT;
1424         int seg;
1425         int offset = offset_in_page(gpa);
1426         int ret;
1427
1428         while ((seg = next_segment(len, offset)) != 0) {
1429                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1430                 if (ret < 0)
1431                         return ret;
1432                 offset = 0;
1433                 len -= seg;
1434                 data += seg;
1435                 ++gfn;
1436         }
1437         return 0;
1438 }
1439
1440 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1441                               gpa_t gpa)
1442 {
1443         struct kvm_memslots *slots = kvm_memslots(kvm);
1444         int offset = offset_in_page(gpa);
1445         gfn_t gfn = gpa >> PAGE_SHIFT;
1446
1447         ghc->gpa = gpa;
1448         ghc->generation = slots->generation;
1449         ghc->memslot = gfn_to_memslot(kvm, gfn);
1450         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1451         if (!kvm_is_error_hva(ghc->hva))
1452                 ghc->hva += offset;
1453         else
1454                 return -EFAULT;
1455
1456         return 0;
1457 }
1458 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1459
1460 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1461                            void *data, unsigned long len)
1462 {
1463         struct kvm_memslots *slots = kvm_memslots(kvm);
1464         int r;
1465
1466         if (slots->generation != ghc->generation)
1467                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1468
1469         if (kvm_is_error_hva(ghc->hva))
1470                 return -EFAULT;
1471
1472         r = __copy_to_user((void __user *)ghc->hva, data, len);
1473         if (r)
1474                 return -EFAULT;
1475         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1476
1477         return 0;
1478 }
1479 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1480
1481 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1482                            void *data, unsigned long len)
1483 {
1484         struct kvm_memslots *slots = kvm_memslots(kvm);
1485         int r;
1486
1487         if (slots->generation != ghc->generation)
1488                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1489
1490         if (kvm_is_error_hva(ghc->hva))
1491                 return -EFAULT;
1492
1493         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1494         if (r)
1495                 return -EFAULT;
1496
1497         return 0;
1498 }
1499 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1500
1501 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1502 {
1503         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1504                                     offset, len);
1505 }
1506 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1507
1508 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1509 {
1510         gfn_t gfn = gpa >> PAGE_SHIFT;
1511         int seg;
1512         int offset = offset_in_page(gpa);
1513         int ret;
1514
1515         while ((seg = next_segment(len, offset)) != 0) {
1516                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1517                 if (ret < 0)
1518                         return ret;
1519                 offset = 0;
1520                 len -= seg;
1521                 ++gfn;
1522         }
1523         return 0;
1524 }
1525 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1526
1527 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1528                              gfn_t gfn)
1529 {
1530         if (memslot && memslot->dirty_bitmap) {
1531                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1532
1533                 /* TODO: introduce set_bit_le() and use it */
1534                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1535         }
1536 }
1537
1538 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1539 {
1540         struct kvm_memory_slot *memslot;
1541
1542         memslot = gfn_to_memslot(kvm, gfn);
1543         mark_page_dirty_in_slot(kvm, memslot, gfn);
1544 }
1545
1546 /*
1547  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1548  */
1549 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1550 {
1551         DEFINE_WAIT(wait);
1552
1553         for (;;) {
1554                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1555
1556                 if (kvm_arch_vcpu_runnable(vcpu)) {
1557                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1558                         break;
1559                 }
1560                 if (kvm_cpu_has_pending_timer(vcpu))
1561                         break;
1562                 if (signal_pending(current))
1563                         break;
1564
1565                 schedule();
1566         }
1567
1568         finish_wait(&vcpu->wq, &wait);
1569 }
1570
1571 #ifndef CONFIG_S390
1572 /*
1573  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1574  */
1575 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1576 {
1577         int me;
1578         int cpu = vcpu->cpu;
1579         wait_queue_head_t *wqp;
1580
1581         wqp = kvm_arch_vcpu_wq(vcpu);
1582         if (waitqueue_active(wqp)) {
1583                 wake_up_interruptible(wqp);
1584                 ++vcpu->stat.halt_wakeup;
1585         }
1586
1587         me = get_cpu();
1588         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1589                 if (kvm_arch_vcpu_should_kick(vcpu))
1590                         smp_send_reschedule(cpu);
1591         put_cpu();
1592 }
1593 #endif /* !CONFIG_S390 */
1594
1595 void kvm_resched(struct kvm_vcpu *vcpu)
1596 {
1597         if (!need_resched())
1598                 return;
1599         cond_resched();
1600 }
1601 EXPORT_SYMBOL_GPL(kvm_resched);
1602
1603 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1604 {
1605         struct pid *pid;
1606         struct task_struct *task = NULL;
1607
1608         rcu_read_lock();
1609         pid = rcu_dereference(target->pid);
1610         if (pid)
1611                 task = get_pid_task(target->pid, PIDTYPE_PID);
1612         rcu_read_unlock();
1613         if (!task)
1614                 return false;
1615         if (task->flags & PF_VCPU) {
1616                 put_task_struct(task);
1617                 return false;
1618         }
1619         if (yield_to(task, 1)) {
1620                 put_task_struct(task);
1621                 return true;
1622         }
1623         put_task_struct(task);
1624         return false;
1625 }
1626 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1627
1628 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1629 /*
1630  * Helper that checks whether a VCPU is eligible for directed yield.
1631  * Most eligible candidate to yield is decided by following heuristics:
1632  *
1633  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1634  *  (preempted lock holder), indicated by @in_spin_loop.
1635  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1636  *
1637  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1638  *  chance last time (mostly it has become eligible now since we have probably
1639  *  yielded to lockholder in last iteration. This is done by toggling
1640  *  @dy_eligible each time a VCPU checked for eligibility.)
1641  *
1642  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1643  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1644  *  burning. Giving priority for a potential lock-holder increases lock
1645  *  progress.
1646  *
1647  *  Since algorithm is based on heuristics, accessing another VCPU data without
1648  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1649  *  and continue with next VCPU and so on.
1650  */
1651 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1652 {
1653         bool eligible;
1654
1655         eligible = !vcpu->spin_loop.in_spin_loop ||
1656                         (vcpu->spin_loop.in_spin_loop &&
1657                          vcpu->spin_loop.dy_eligible);
1658
1659         if (vcpu->spin_loop.in_spin_loop)
1660                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1661
1662         return eligible;
1663 }
1664 #endif
1665 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1666 {
1667         struct kvm *kvm = me->kvm;
1668         struct kvm_vcpu *vcpu;
1669         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1670         int yielded = 0;
1671         int pass;
1672         int i;
1673
1674         kvm_vcpu_set_in_spin_loop(me, true);
1675         /*
1676          * We boost the priority of a VCPU that is runnable but not
1677          * currently running, because it got preempted by something
1678          * else and called schedule in __vcpu_run.  Hopefully that
1679          * VCPU is holding the lock that we need and will release it.
1680          * We approximate round-robin by starting at the last boosted VCPU.
1681          */
1682         for (pass = 0; pass < 2 && !yielded; pass++) {
1683                 kvm_for_each_vcpu(i, vcpu, kvm) {
1684                         if (!pass && i <= last_boosted_vcpu) {
1685                                 i = last_boosted_vcpu;
1686                                 continue;
1687                         } else if (pass && i > last_boosted_vcpu)
1688                                 break;
1689                         if (vcpu == me)
1690                                 continue;
1691                         if (waitqueue_active(&vcpu->wq))
1692                                 continue;
1693                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1694                                 continue;
1695                         if (kvm_vcpu_yield_to(vcpu)) {
1696                                 kvm->last_boosted_vcpu = i;
1697                                 yielded = 1;
1698                                 break;
1699                         }
1700                 }
1701         }
1702         kvm_vcpu_set_in_spin_loop(me, false);
1703
1704         /* Ensure vcpu is not eligible during next spinloop */
1705         kvm_vcpu_set_dy_eligible(me, false);
1706 }
1707 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1708
1709 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1710 {
1711         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1712         struct page *page;
1713
1714         if (vmf->pgoff == 0)
1715                 page = virt_to_page(vcpu->run);
1716 #ifdef CONFIG_X86
1717         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1718                 page = virt_to_page(vcpu->arch.pio_data);
1719 #endif
1720 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1721         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1722                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1723 #endif
1724         else
1725                 return kvm_arch_vcpu_fault(vcpu, vmf);
1726         get_page(page);
1727         vmf->page = page;
1728         return 0;
1729 }
1730
1731 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1732         .fault = kvm_vcpu_fault,
1733 };
1734
1735 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1736 {
1737         vma->vm_ops = &kvm_vcpu_vm_ops;
1738         return 0;
1739 }
1740
1741 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1742 {
1743         struct kvm_vcpu *vcpu = filp->private_data;
1744
1745         kvm_put_kvm(vcpu->kvm);
1746         return 0;
1747 }
1748
1749 static struct file_operations kvm_vcpu_fops = {
1750         .release        = kvm_vcpu_release,
1751         .unlocked_ioctl = kvm_vcpu_ioctl,
1752 #ifdef CONFIG_COMPAT
1753         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1754 #endif
1755         .mmap           = kvm_vcpu_mmap,
1756         .llseek         = noop_llseek,
1757 };
1758
1759 /*
1760  * Allocates an inode for the vcpu.
1761  */
1762 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1763 {
1764         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1765 }
1766
1767 /*
1768  * Creates some virtual cpus.  Good luck creating more than one.
1769  */
1770 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1771 {
1772         int r;
1773         struct kvm_vcpu *vcpu, *v;
1774
1775         vcpu = kvm_arch_vcpu_create(kvm, id);
1776         if (IS_ERR(vcpu))
1777                 return PTR_ERR(vcpu);
1778
1779         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1780
1781         r = kvm_arch_vcpu_setup(vcpu);
1782         if (r)
1783                 goto vcpu_destroy;
1784
1785         mutex_lock(&kvm->lock);
1786         if (!kvm_vcpu_compatible(vcpu)) {
1787                 r = -EINVAL;
1788                 goto unlock_vcpu_destroy;
1789         }
1790         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1791                 r = -EINVAL;
1792                 goto unlock_vcpu_destroy;
1793         }
1794
1795         kvm_for_each_vcpu(r, v, kvm)
1796                 if (v->vcpu_id == id) {
1797                         r = -EEXIST;
1798                         goto unlock_vcpu_destroy;
1799                 }
1800
1801         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1802
1803         /* Now it's all set up, let userspace reach it */
1804         kvm_get_kvm(kvm);
1805         r = create_vcpu_fd(vcpu);
1806         if (r < 0) {
1807                 kvm_put_kvm(kvm);
1808                 goto unlock_vcpu_destroy;
1809         }
1810
1811         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1812         smp_wmb();
1813         atomic_inc(&kvm->online_vcpus);
1814
1815         mutex_unlock(&kvm->lock);
1816         return r;
1817
1818 unlock_vcpu_destroy:
1819         mutex_unlock(&kvm->lock);
1820 vcpu_destroy:
1821         kvm_arch_vcpu_destroy(vcpu);
1822         return r;
1823 }
1824
1825 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1826 {
1827         if (sigset) {
1828                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1829                 vcpu->sigset_active = 1;
1830                 vcpu->sigset = *sigset;
1831         } else
1832                 vcpu->sigset_active = 0;
1833         return 0;
1834 }
1835
1836 static long kvm_vcpu_ioctl(struct file *filp,
1837                            unsigned int ioctl, unsigned long arg)
1838 {
1839         struct kvm_vcpu *vcpu = filp->private_data;
1840         void __user *argp = (void __user *)arg;
1841         int r;
1842         struct kvm_fpu *fpu = NULL;
1843         struct kvm_sregs *kvm_sregs = NULL;
1844
1845         if (vcpu->kvm->mm != current->mm)
1846                 return -EIO;
1847
1848 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1849         /*
1850          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1851          * so vcpu_load() would break it.
1852          */
1853         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1854                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1855 #endif
1856
1857
1858         vcpu_load(vcpu);
1859         switch (ioctl) {
1860         case KVM_RUN:
1861                 r = -EINVAL;
1862                 if (arg)
1863                         goto out;
1864                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1865                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1866                 break;
1867         case KVM_GET_REGS: {
1868                 struct kvm_regs *kvm_regs;
1869
1870                 r = -ENOMEM;
1871                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1872                 if (!kvm_regs)
1873                         goto out;
1874                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1875                 if (r)
1876                         goto out_free1;
1877                 r = -EFAULT;
1878                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1879                         goto out_free1;
1880                 r = 0;
1881 out_free1:
1882                 kfree(kvm_regs);
1883                 break;
1884         }
1885         case KVM_SET_REGS: {
1886                 struct kvm_regs *kvm_regs;
1887
1888                 r = -ENOMEM;
1889                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1890                 if (IS_ERR(kvm_regs)) {
1891                         r = PTR_ERR(kvm_regs);
1892                         goto out;
1893                 }
1894                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1895                 if (r)
1896                         goto out_free2;
1897                 r = 0;
1898 out_free2:
1899                 kfree(kvm_regs);
1900                 break;
1901         }
1902         case KVM_GET_SREGS: {
1903                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1904                 r = -ENOMEM;
1905                 if (!kvm_sregs)
1906                         goto out;
1907                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1908                 if (r)
1909                         goto out;
1910                 r = -EFAULT;
1911                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1912                         goto out;
1913                 r = 0;
1914                 break;
1915         }
1916         case KVM_SET_SREGS: {
1917                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1918                 if (IS_ERR(kvm_sregs)) {
1919                         r = PTR_ERR(kvm_sregs);
1920                         goto out;
1921                 }
1922                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1923                 if (r)
1924                         goto out;
1925                 r = 0;
1926                 break;
1927         }
1928         case KVM_GET_MP_STATE: {
1929                 struct kvm_mp_state mp_state;
1930
1931                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1932                 if (r)
1933                         goto out;
1934                 r = -EFAULT;
1935                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1936                         goto out;
1937                 r = 0;
1938                 break;
1939         }
1940         case KVM_SET_MP_STATE: {
1941                 struct kvm_mp_state mp_state;
1942
1943                 r = -EFAULT;
1944                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1945                         goto out;
1946                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1947                 if (r)
1948                         goto out;
1949                 r = 0;
1950                 break;
1951         }
1952         case KVM_TRANSLATE: {
1953                 struct kvm_translation tr;
1954
1955                 r = -EFAULT;
1956                 if (copy_from_user(&tr, argp, sizeof tr))
1957                         goto out;
1958                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1959                 if (r)
1960                         goto out;
1961                 r = -EFAULT;
1962                 if (copy_to_user(argp, &tr, sizeof tr))
1963                         goto out;
1964                 r = 0;
1965                 break;
1966         }
1967         case KVM_SET_GUEST_DEBUG: {
1968                 struct kvm_guest_debug dbg;
1969
1970                 r = -EFAULT;
1971                 if (copy_from_user(&dbg, argp, sizeof dbg))
1972                         goto out;
1973                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1974                 if (r)
1975                         goto out;
1976                 r = 0;
1977                 break;
1978         }
1979         case KVM_SET_SIGNAL_MASK: {
1980                 struct kvm_signal_mask __user *sigmask_arg = argp;
1981                 struct kvm_signal_mask kvm_sigmask;
1982                 sigset_t sigset, *p;
1983
1984                 p = NULL;
1985                 if (argp) {
1986                         r = -EFAULT;
1987                         if (copy_from_user(&kvm_sigmask, argp,
1988                                            sizeof kvm_sigmask))
1989                                 goto out;
1990                         r = -EINVAL;
1991                         if (kvm_sigmask.len != sizeof sigset)
1992                                 goto out;
1993                         r = -EFAULT;
1994                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1995                                            sizeof sigset))
1996                                 goto out;
1997                         p = &sigset;
1998                 }
1999                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2000                 break;
2001         }
2002         case KVM_GET_FPU: {
2003                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2004                 r = -ENOMEM;
2005                 if (!fpu)
2006                         goto out;
2007                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2008                 if (r)
2009                         goto out;
2010                 r = -EFAULT;
2011                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2012                         goto out;
2013                 r = 0;
2014                 break;
2015         }
2016         case KVM_SET_FPU: {
2017                 fpu = memdup_user(argp, sizeof(*fpu));
2018                 if (IS_ERR(fpu)) {
2019                         r = PTR_ERR(fpu);
2020                         goto out;
2021                 }
2022                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2023                 if (r)
2024                         goto out;
2025                 r = 0;
2026                 break;
2027         }
2028         default:
2029                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2030         }
2031 out:
2032         vcpu_put(vcpu);
2033         kfree(fpu);
2034         kfree(kvm_sregs);
2035         return r;
2036 }
2037
2038 #ifdef CONFIG_COMPAT
2039 static long kvm_vcpu_compat_ioctl(struct file *filp,
2040                                   unsigned int ioctl, unsigned long arg)
2041 {
2042         struct kvm_vcpu *vcpu = filp->private_data;
2043         void __user *argp = compat_ptr(arg);
2044         int r;
2045
2046         if (vcpu->kvm->mm != current->mm)
2047                 return -EIO;
2048
2049         switch (ioctl) {
2050         case KVM_SET_SIGNAL_MASK: {
2051                 struct kvm_signal_mask __user *sigmask_arg = argp;
2052                 struct kvm_signal_mask kvm_sigmask;
2053                 compat_sigset_t csigset;
2054                 sigset_t sigset;
2055
2056                 if (argp) {
2057                         r = -EFAULT;
2058                         if (copy_from_user(&kvm_sigmask, argp,
2059                                            sizeof kvm_sigmask))
2060                                 goto out;
2061                         r = -EINVAL;
2062                         if (kvm_sigmask.len != sizeof csigset)
2063                                 goto out;
2064                         r = -EFAULT;
2065                         if (copy_from_user(&csigset, sigmask_arg->sigset,
2066                                            sizeof csigset))
2067                                 goto out;
2068                 }
2069                 sigset_from_compat(&sigset, &csigset);
2070                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2071                 break;
2072         }
2073         default:
2074                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
2075         }
2076
2077 out:
2078         return r;
2079 }
2080 #endif
2081
2082 static long kvm_vm_ioctl(struct file *filp,
2083                            unsigned int ioctl, unsigned long arg)
2084 {
2085         struct kvm *kvm = filp->private_data;
2086         void __user *argp = (void __user *)arg;
2087         int r;
2088
2089         if (kvm->mm != current->mm)
2090                 return -EIO;
2091         switch (ioctl) {
2092         case KVM_CREATE_VCPU:
2093                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2094                 if (r < 0)
2095                         goto out;
2096                 break;
2097         case KVM_SET_USER_MEMORY_REGION: {
2098                 struct kvm_userspace_memory_region kvm_userspace_mem;
2099
2100                 r = -EFAULT;
2101                 if (copy_from_user(&kvm_userspace_mem, argp,
2102                                                 sizeof kvm_userspace_mem))
2103                         goto out;
2104
2105                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2106                 if (r)
2107                         goto out;
2108                 break;
2109         }
2110         case KVM_GET_DIRTY_LOG: {
2111                 struct kvm_dirty_log log;
2112
2113                 r = -EFAULT;
2114                 if (copy_from_user(&log, argp, sizeof log))
2115                         goto out;
2116                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2117                 if (r)
2118                         goto out;
2119                 break;
2120         }
2121 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2122         case KVM_REGISTER_COALESCED_MMIO: {
2123                 struct kvm_coalesced_mmio_zone zone;
2124                 r = -EFAULT;
2125                 if (copy_from_user(&zone, argp, sizeof zone))
2126                         goto out;
2127                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2128                 if (r)
2129                         goto out;
2130                 r = 0;
2131                 break;
2132         }
2133         case KVM_UNREGISTER_COALESCED_MMIO: {
2134                 struct kvm_coalesced_mmio_zone zone;
2135                 r = -EFAULT;
2136                 if (copy_from_user(&zone, argp, sizeof zone))
2137                         goto out;
2138                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2139                 if (r)
2140                         goto out;
2141                 r = 0;
2142                 break;
2143         }
2144 #endif
2145         case KVM_IRQFD: {
2146                 struct kvm_irqfd data;
2147
2148                 r = -EFAULT;
2149                 if (copy_from_user(&data, argp, sizeof data))
2150                         goto out;
2151                 r = kvm_irqfd(kvm, &data);
2152                 break;
2153         }
2154         case KVM_IOEVENTFD: {
2155                 struct kvm_ioeventfd data;
2156
2157                 r = -EFAULT;
2158                 if (copy_from_user(&data, argp, sizeof data))
2159                         goto out;
2160                 r = kvm_ioeventfd(kvm, &data);
2161                 break;
2162         }
2163 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2164         case KVM_SET_BOOT_CPU_ID:
2165                 r = 0;
2166                 mutex_lock(&kvm->lock);
2167                 if (atomic_read(&kvm->online_vcpus) != 0)
2168                         r = -EBUSY;
2169                 else
2170                         kvm->bsp_vcpu_id = arg;
2171                 mutex_unlock(&kvm->lock);
2172                 break;
2173 #endif
2174 #ifdef CONFIG_HAVE_KVM_MSI
2175         case KVM_SIGNAL_MSI: {
2176                 struct kvm_msi msi;
2177
2178                 r = -EFAULT;
2179                 if (copy_from_user(&msi, argp, sizeof msi))
2180                         goto out;
2181                 r = kvm_send_userspace_msi(kvm, &msi);
2182                 break;
2183         }
2184 #endif
2185 #ifdef __KVM_HAVE_IRQ_LINE
2186         case KVM_IRQ_LINE_STATUS:
2187         case KVM_IRQ_LINE: {
2188                 struct kvm_irq_level irq_event;
2189
2190                 r = -EFAULT;
2191                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2192                         goto out;
2193
2194                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2195                 if (r)
2196                         goto out;
2197
2198                 r = -EFAULT;
2199                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2200                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2201                                 goto out;
2202                 }
2203
2204                 r = 0;
2205                 break;
2206         }
2207 #endif
2208         default:
2209                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2210                 if (r == -ENOTTY)
2211                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2212         }
2213 out:
2214         return r;
2215 }
2216
2217 #ifdef CONFIG_COMPAT
2218 struct compat_kvm_dirty_log {
2219         __u32 slot;
2220         __u32 padding1;
2221         union {
2222                 compat_uptr_t dirty_bitmap; /* one bit per page */
2223                 __u64 padding2;
2224         };
2225 };
2226
2227 static long kvm_vm_compat_ioctl(struct file *filp,
2228                            unsigned int ioctl, unsigned long arg)
2229 {
2230         struct kvm *kvm = filp->private_data;
2231         int r;
2232
2233         if (kvm->mm != current->mm)
2234                 return -EIO;
2235         switch (ioctl) {
2236         case KVM_GET_DIRTY_LOG: {
2237                 struct compat_kvm_dirty_log compat_log;
2238                 struct kvm_dirty_log log;
2239
2240                 r = -EFAULT;
2241                 if (copy_from_user(&compat_log, (void __user *)arg,
2242                                    sizeof(compat_log)))
2243                         goto out;
2244                 log.slot         = compat_log.slot;
2245                 log.padding1     = compat_log.padding1;
2246                 log.padding2     = compat_log.padding2;
2247                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2248
2249                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2250                 if (r)
2251                         goto out;
2252                 break;
2253         }
2254         default:
2255                 r = kvm_vm_ioctl(filp, ioctl, arg);
2256         }
2257
2258 out:
2259         return r;
2260 }
2261 #endif
2262
2263 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2264 {
2265         struct page *page[1];
2266         unsigned long addr;
2267         int npages;
2268         gfn_t gfn = vmf->pgoff;
2269         struct kvm *kvm = vma->vm_file->private_data;
2270
2271         addr = gfn_to_hva(kvm, gfn);
2272         if (kvm_is_error_hva(addr))
2273                 return VM_FAULT_SIGBUS;
2274
2275         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2276                                 NULL);
2277         if (unlikely(npages != 1))
2278                 return VM_FAULT_SIGBUS;
2279
2280         vmf->page = page[0];
2281         return 0;
2282 }
2283
2284 static const struct vm_operations_struct kvm_vm_vm_ops = {
2285         .fault = kvm_vm_fault,
2286 };
2287
2288 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2289 {
2290         vma->vm_ops = &kvm_vm_vm_ops;
2291         return 0;
2292 }
2293
2294 static struct file_operations kvm_vm_fops = {
2295         .release        = kvm_vm_release,
2296         .unlocked_ioctl = kvm_vm_ioctl,
2297 #ifdef CONFIG_COMPAT
2298         .compat_ioctl   = kvm_vm_compat_ioctl,
2299 #endif
2300         .mmap           = kvm_vm_mmap,
2301         .llseek         = noop_llseek,
2302 };
2303
2304 static int kvm_dev_ioctl_create_vm(unsigned long type)
2305 {
2306         int r;
2307         struct kvm *kvm;
2308
2309         kvm = kvm_create_vm(type);
2310         if (IS_ERR(kvm))
2311                 return PTR_ERR(kvm);
2312 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2313         r = kvm_coalesced_mmio_init(kvm);
2314         if (r < 0) {
2315                 kvm_put_kvm(kvm);
2316                 return r;
2317         }
2318 #endif
2319         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2320         if (r < 0)
2321                 kvm_put_kvm(kvm);
2322
2323         return r;
2324 }
2325
2326 static long kvm_dev_ioctl_check_extension_generic(long arg)
2327 {
2328         switch (arg) {
2329         case KVM_CAP_USER_MEMORY:
2330         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2331         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2332 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2333         case KVM_CAP_SET_BOOT_CPU_ID:
2334 #endif
2335         case KVM_CAP_INTERNAL_ERROR_DATA:
2336 #ifdef CONFIG_HAVE_KVM_MSI
2337         case KVM_CAP_SIGNAL_MSI:
2338 #endif
2339                 return 1;
2340 #ifdef KVM_CAP_IRQ_ROUTING
2341         case KVM_CAP_IRQ_ROUTING:
2342                 return KVM_MAX_IRQ_ROUTES;
2343 #endif
2344         default:
2345                 break;
2346         }
2347         return kvm_dev_ioctl_check_extension(arg);
2348 }
2349
2350 static long kvm_dev_ioctl(struct file *filp,
2351                           unsigned int ioctl, unsigned long arg)
2352 {
2353         long r = -EINVAL;
2354
2355         switch (ioctl) {
2356         case KVM_GET_API_VERSION:
2357                 r = -EINVAL;
2358                 if (arg)
2359                         goto out;
2360                 r = KVM_API_VERSION;
2361                 break;
2362         case KVM_CREATE_VM:
2363                 r = kvm_dev_ioctl_create_vm(arg);
2364                 break;
2365         case KVM_CHECK_EXTENSION:
2366                 r = kvm_dev_ioctl_check_extension_generic(arg);
2367                 break;
2368         case KVM_GET_VCPU_MMAP_SIZE:
2369                 r = -EINVAL;
2370                 if (arg)
2371                         goto out;
2372                 r = PAGE_SIZE;     /* struct kvm_run */
2373 #ifdef CONFIG_X86
2374                 r += PAGE_SIZE;    /* pio data page */
2375 #endif
2376 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2377                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2378 #endif
2379                 break;
2380         case KVM_TRACE_ENABLE:
2381         case KVM_TRACE_PAUSE:
2382         case KVM_TRACE_DISABLE:
2383                 r = -EOPNOTSUPP;
2384                 break;
2385         default:
2386                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2387         }
2388 out:
2389         return r;
2390 }
2391
2392 static struct file_operations kvm_chardev_ops = {
2393         .unlocked_ioctl = kvm_dev_ioctl,
2394         .compat_ioctl   = kvm_dev_ioctl,
2395         .llseek         = noop_llseek,
2396 };
2397
2398 static struct miscdevice kvm_dev = {
2399         KVM_MINOR,
2400         "kvm",
2401         &kvm_chardev_ops,
2402 };
2403
2404 static void hardware_enable_nolock(void *junk)
2405 {
2406         int cpu = raw_smp_processor_id();
2407         int r;
2408
2409         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2410                 return;
2411
2412         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2413
2414         r = kvm_arch_hardware_enable(NULL);
2415
2416         if (r) {
2417                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2418                 atomic_inc(&hardware_enable_failed);
2419                 printk(KERN_INFO "kvm: enabling virtualization on "
2420                                  "CPU%d failed\n", cpu);
2421         }
2422 }
2423
2424 static void hardware_enable(void *junk)
2425 {
2426         raw_spin_lock(&kvm_lock);
2427         hardware_enable_nolock(junk);
2428         raw_spin_unlock(&kvm_lock);
2429 }
2430
2431 static void hardware_disable_nolock(void *junk)
2432 {
2433         int cpu = raw_smp_processor_id();
2434
2435         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2436                 return;
2437         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2438         kvm_arch_hardware_disable(NULL);
2439 }
2440
2441 static void hardware_disable(void *junk)
2442 {
2443         raw_spin_lock(&kvm_lock);
2444         hardware_disable_nolock(junk);
2445         raw_spin_unlock(&kvm_lock);
2446 }
2447
2448 static void hardware_disable_all_nolock(void)
2449 {
2450         BUG_ON(!kvm_usage_count);
2451
2452         kvm_usage_count--;
2453         if (!kvm_usage_count)
2454                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2455 }
2456
2457 static void hardware_disable_all(void)
2458 {
2459         raw_spin_lock(&kvm_lock);
2460         hardware_disable_all_nolock();
2461         raw_spin_unlock(&kvm_lock);
2462 }
2463
2464 static int hardware_enable_all(void)
2465 {
2466         int r = 0;
2467
2468         raw_spin_lock(&kvm_lock);
2469
2470         kvm_usage_count++;
2471         if (kvm_usage_count == 1) {
2472                 atomic_set(&hardware_enable_failed, 0);
2473                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2474
2475                 if (atomic_read(&hardware_enable_failed)) {
2476                         hardware_disable_all_nolock();
2477                         r = -EBUSY;
2478                 }
2479         }
2480
2481         raw_spin_unlock(&kvm_lock);
2482
2483         return r;
2484 }
2485
2486 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2487                            void *v)
2488 {
2489         int cpu = (long)v;
2490
2491         if (!kvm_usage_count)
2492                 return NOTIFY_OK;
2493
2494         val &= ~CPU_TASKS_FROZEN;
2495         switch (val) {
2496         case CPU_DYING:
2497                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2498                        cpu);
2499                 hardware_disable(NULL);
2500                 break;
2501         case CPU_STARTING:
2502                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2503                        cpu);
2504                 hardware_enable(NULL);
2505                 break;
2506         }
2507         return NOTIFY_OK;
2508 }
2509
2510
2511 asmlinkage void kvm_spurious_fault(void)
2512 {
2513         /* Fault while not rebooting.  We want the trace. */
2514         BUG();
2515 }
2516 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2517
2518 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2519                       void *v)
2520 {
2521         /*
2522          * Some (well, at least mine) BIOSes hang on reboot if
2523          * in vmx root mode.
2524          *
2525          * And Intel TXT required VMX off for all cpu when system shutdown.
2526          */
2527         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2528         kvm_rebooting = true;
2529         on_each_cpu(hardware_disable_nolock, NULL, 1);
2530         return NOTIFY_OK;
2531 }
2532
2533 static struct notifier_block kvm_reboot_notifier = {
2534         .notifier_call = kvm_reboot,
2535         .priority = 0,
2536 };
2537
2538 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2539 {
2540         int i;
2541
2542         for (i = 0; i < bus->dev_count; i++) {
2543                 struct kvm_io_device *pos = bus->range[i].dev;
2544
2545                 kvm_iodevice_destructor(pos);
2546         }
2547         kfree(bus);
2548 }
2549
2550 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2551 {
2552         const struct kvm_io_range *r1 = p1;
2553         const struct kvm_io_range *r2 = p2;
2554
2555         if (r1->addr < r2->addr)
2556                 return -1;
2557         if (r1->addr + r1->len > r2->addr + r2->len)
2558                 return 1;
2559         return 0;
2560 }
2561
2562 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2563                           gpa_t addr, int len)
2564 {
2565         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2566                 .addr = addr,
2567                 .len = len,
2568                 .dev = dev,
2569         };
2570
2571         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2572                 kvm_io_bus_sort_cmp, NULL);
2573
2574         return 0;
2575 }
2576
2577 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2578                              gpa_t addr, int len)
2579 {
2580         struct kvm_io_range *range, key;
2581         int off;
2582
2583         key = (struct kvm_io_range) {
2584                 .addr = addr,
2585                 .len = len,
2586         };
2587
2588         range = bsearch(&key, bus->range, bus->dev_count,
2589                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2590         if (range == NULL)
2591                 return -ENOENT;
2592
2593         off = range - bus->range;
2594
2595         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2596                 off--;
2597
2598         return off;
2599 }
2600
2601 /* kvm_io_bus_write - called under kvm->slots_lock */
2602 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2603                      int len, const void *val)
2604 {
2605         int idx;
2606         struct kvm_io_bus *bus;
2607         struct kvm_io_range range;
2608
2609         range = (struct kvm_io_range) {
2610                 .addr = addr,
2611                 .len = len,
2612         };
2613
2614         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2615         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2616         if (idx < 0)
2617                 return -EOPNOTSUPP;
2618
2619         while (idx < bus->dev_count &&
2620                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2621                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2622                         return 0;
2623                 idx++;
2624         }
2625
2626         return -EOPNOTSUPP;
2627 }
2628
2629 /* kvm_io_bus_read - called under kvm->slots_lock */
2630 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2631                     int len, void *val)
2632 {
2633         int idx;
2634         struct kvm_io_bus *bus;
2635         struct kvm_io_range range;
2636
2637         range = (struct kvm_io_range) {
2638                 .addr = addr,
2639                 .len = len,
2640         };
2641
2642         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2643         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2644         if (idx < 0)
2645                 return -EOPNOTSUPP;
2646
2647         while (idx < bus->dev_count &&
2648                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2649                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2650                         return 0;
2651                 idx++;
2652         }
2653
2654         return -EOPNOTSUPP;
2655 }
2656
2657 /* Caller must hold slots_lock. */
2658 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2659                             int len, struct kvm_io_device *dev)
2660 {
2661         struct kvm_io_bus *new_bus, *bus;
2662
2663         bus = kvm->buses[bus_idx];
2664         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2665                 return -ENOSPC;
2666
2667         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2668                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2669         if (!new_bus)
2670                 return -ENOMEM;
2671         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2672                sizeof(struct kvm_io_range)));
2673         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2674         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2675         synchronize_srcu_expedited(&kvm->srcu);
2676         kfree(bus);
2677
2678         return 0;
2679 }
2680
2681 /* Caller must hold slots_lock. */
2682 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2683                               struct kvm_io_device *dev)
2684 {
2685         int i, r;
2686         struct kvm_io_bus *new_bus, *bus;
2687
2688         bus = kvm->buses[bus_idx];
2689         r = -ENOENT;
2690         for (i = 0; i < bus->dev_count; i++)
2691                 if (bus->range[i].dev == dev) {
2692                         r = 0;
2693                         break;
2694                 }
2695
2696         if (r)
2697                 return r;
2698
2699         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2700                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2701         if (!new_bus)
2702                 return -ENOMEM;
2703
2704         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2705         new_bus->dev_count--;
2706         memcpy(new_bus->range + i, bus->range + i + 1,
2707                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2708
2709         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2710         synchronize_srcu_expedited(&kvm->srcu);
2711         kfree(bus);
2712         return r;
2713 }
2714
2715 static struct notifier_block kvm_cpu_notifier = {
2716         .notifier_call = kvm_cpu_hotplug,
2717 };
2718
2719 static int vm_stat_get(void *_offset, u64 *val)
2720 {
2721         unsigned offset = (long)_offset;
2722         struct kvm *kvm;
2723
2724         *val = 0;
2725         raw_spin_lock(&kvm_lock);
2726         list_for_each_entry(kvm, &vm_list, vm_list)
2727                 *val += *(u32 *)((void *)kvm + offset);
2728         raw_spin_unlock(&kvm_lock);
2729         return 0;
2730 }
2731
2732 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2733
2734 static int vcpu_stat_get(void *_offset, u64 *val)
2735 {
2736         unsigned offset = (long)_offset;
2737         struct kvm *kvm;
2738         struct kvm_vcpu *vcpu;
2739         int i;
2740
2741         *val = 0;
2742         raw_spin_lock(&kvm_lock);
2743         list_for_each_entry(kvm, &vm_list, vm_list)
2744                 kvm_for_each_vcpu(i, vcpu, kvm)
2745                         *val += *(u32 *)((void *)vcpu + offset);
2746
2747         raw_spin_unlock(&kvm_lock);
2748         return 0;
2749 }
2750
2751 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2752
2753 static const struct file_operations *stat_fops[] = {
2754         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2755         [KVM_STAT_VM]   = &vm_stat_fops,
2756 };
2757
2758 static int kvm_init_debug(void)
2759 {
2760         int r = -EFAULT;
2761         struct kvm_stats_debugfs_item *p;
2762
2763         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2764         if (kvm_debugfs_dir == NULL)
2765                 goto out;
2766
2767         for (p = debugfs_entries; p->name; ++p) {
2768                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2769                                                 (void *)(long)p->offset,
2770                                                 stat_fops[p->kind]);
2771                 if (p->dentry == NULL)
2772                         goto out_dir;
2773         }
2774
2775         return 0;
2776
2777 out_dir:
2778         debugfs_remove_recursive(kvm_debugfs_dir);
2779 out:
2780         return r;
2781 }
2782
2783 static void kvm_exit_debug(void)
2784 {
2785         struct kvm_stats_debugfs_item *p;
2786
2787         for (p = debugfs_entries; p->name; ++p)
2788                 debugfs_remove(p->dentry);
2789         debugfs_remove(kvm_debugfs_dir);
2790 }
2791
2792 static int kvm_suspend(void)
2793 {
2794         if (kvm_usage_count)
2795                 hardware_disable_nolock(NULL);
2796         return 0;
2797 }
2798
2799 static void kvm_resume(void)
2800 {
2801         if (kvm_usage_count) {
2802                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2803                 hardware_enable_nolock(NULL);
2804         }
2805 }
2806
2807 static struct syscore_ops kvm_syscore_ops = {
2808         .suspend = kvm_suspend,
2809         .resume = kvm_resume,
2810 };
2811
2812 static inline
2813 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2814 {
2815         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2816 }
2817
2818 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2819 {
2820         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2821
2822         kvm_arch_vcpu_load(vcpu, cpu);
2823 }
2824
2825 static void kvm_sched_out(struct preempt_notifier *pn,
2826                           struct task_struct *next)
2827 {
2828         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2829
2830         kvm_arch_vcpu_put(vcpu);
2831 }
2832
2833 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2834                   struct module *module)
2835 {
2836         int r;
2837         int cpu;
2838
2839         r = kvm_arch_init(opaque);
2840         if (r)
2841                 goto out_fail;
2842
2843         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2844                 r = -ENOMEM;
2845                 goto out_free_0;
2846         }
2847
2848         r = kvm_arch_hardware_setup();
2849         if (r < 0)
2850                 goto out_free_0a;
2851
2852         for_each_online_cpu(cpu) {
2853                 smp_call_function_single(cpu,
2854                                 kvm_arch_check_processor_compat,
2855                                 &r, 1);
2856                 if (r < 0)
2857                         goto out_free_1;
2858         }
2859
2860         r = register_cpu_notifier(&kvm_cpu_notifier);
2861         if (r)
2862                 goto out_free_2;
2863         register_reboot_notifier(&kvm_reboot_notifier);
2864
2865         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2866         if (!vcpu_align)
2867                 vcpu_align = __alignof__(struct kvm_vcpu);
2868         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2869                                            0, NULL);
2870         if (!kvm_vcpu_cache) {
2871                 r = -ENOMEM;
2872                 goto out_free_3;
2873         }
2874
2875         r = kvm_async_pf_init();
2876         if (r)
2877                 goto out_free;
2878
2879         kvm_chardev_ops.owner = module;
2880         kvm_vm_fops.owner = module;
2881         kvm_vcpu_fops.owner = module;
2882
2883         r = misc_register(&kvm_dev);
2884         if (r) {
2885                 printk(KERN_ERR "kvm: misc device register failed\n");
2886                 goto out_unreg;
2887         }
2888
2889         register_syscore_ops(&kvm_syscore_ops);
2890
2891         kvm_preempt_ops.sched_in = kvm_sched_in;
2892         kvm_preempt_ops.sched_out = kvm_sched_out;
2893
2894         r = kvm_init_debug();
2895         if (r) {
2896                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2897                 goto out_undebugfs;
2898         }
2899
2900         return 0;
2901
2902 out_undebugfs:
2903         unregister_syscore_ops(&kvm_syscore_ops);
2904 out_unreg:
2905         kvm_async_pf_deinit();
2906 out_free:
2907         kmem_cache_destroy(kvm_vcpu_cache);
2908 out_free_3:
2909         unregister_reboot_notifier(&kvm_reboot_notifier);
2910         unregister_cpu_notifier(&kvm_cpu_notifier);
2911 out_free_2:
2912 out_free_1:
2913         kvm_arch_hardware_unsetup();
2914 out_free_0a:
2915         free_cpumask_var(cpus_hardware_enabled);
2916 out_free_0:
2917         kvm_arch_exit();
2918 out_fail:
2919         return r;
2920 }
2921 EXPORT_SYMBOL_GPL(kvm_init);
2922
2923 void kvm_exit(void)
2924 {
2925         kvm_exit_debug();
2926         misc_deregister(&kvm_dev);
2927         kmem_cache_destroy(kvm_vcpu_cache);
2928         kvm_async_pf_deinit();
2929         unregister_syscore_ops(&kvm_syscore_ops);
2930         unregister_reboot_notifier(&kvm_reboot_notifier);
2931         unregister_cpu_notifier(&kvm_cpu_notifier);
2932         on_each_cpu(hardware_disable_nolock, NULL, 1);
2933         kvm_arch_hardware_unsetup();
2934         kvm_arch_exit();
2935         free_cpumask_var(cpus_hardware_enabled);
2936 }
2937 EXPORT_SYMBOL_GPL(kvm_exit);