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