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