virt/kvm/kvm_main.c

   1 /*
   2  * Kernel-based Virtual Machine driver for Linux
   3  *
   4  * This module enables machines with Intel VT-x extensions to run virtual
   5  * machines without emulation or binary translation.
   6  *
   7  * Copyright (C) 2006 Qumranet, Inc.
   8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
   9  *
  10  * Authors:
  11  *   Avi Kivity   <avi@qumranet.com>
  12  *   Yaniv Kamay  <yaniv@qumranet.com>
  13  *
  14  * This work is licensed under the terms of the GNU GPL, version 2.  See
  15  * the COPYING file in the top-level directory.
  16  *
  17  */
  18
  19 #include "iodev.h"
  20
  21 #include <linux/kvm_host.h>
  22 #include <linux/kvm.h>
  23 #include <linux/module.h>
  24 #include <linux/errno.h>
  25 #include <linux/percpu.h>
  26 #include <linux/mm.h>
  27 #include <linux/miscdevice.h>
  28 #include <linux/vmalloc.h>
  29 #include <linux/reboot.h>
  30 #include <linux/debugfs.h>
  31 #include <linux/highmem.h>
  32 #include <linux/file.h>
  33 #include <linux/syscore_ops.h>
  34 #include <linux/cpu.h>
  35 #include <linux/sched.h>
  36 #include <linux/cpumask.h>
  37 #include <linux/smp.h>
  38 #include <linux/anon_inodes.h>
  39 #include <linux/profile.h>
  40 #include <linux/kvm_para.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/mman.h>
  43 #include <linux/swap.h>
  44 #include <linux/bitops.h>
  45 #include <linux/spinlock.h>
  46 #include <linux/compat.h>
  47 #include <linux/srcu.h>
  48 #include <linux/hugetlb.h>
  49 #include <linux/slab.h>
  50 #include <linux/sort.h>
  51 #include <linux/bsearch.h>
  52
  53 #include <asm/processor.h>
  54 #include <asm/io.h>
  55 #include <asm/uaccess.h>
  56 #include <asm/pgtable.h>
  57
  58 #include "coalesced_mmio.h"
  59 #include "async_pf.h"
  60
  61 #define CREATE_TRACE_POINTS
  62 #include <trace/events/kvm.h>
  63
  64 MODULE_AUTHOR("Qumranet");
  65 MODULE_LICENSE("GPL");
  66
  67 /*
  68  * Ordering of locks:
  69  *
  70  *              kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  71  */
  72
  73 DEFINE_RAW_SPINLOCK(kvm_lock);
  74 LIST_HEAD(vm_list);
  75
  76 static cpumask_var_t cpus_hardware_enabled;
  77 static int kvm_usage_count = 0;
  78 static atomic_t hardware_enable_failed;
  79
  80 struct kmem_cache *kvm_vcpu_cache;
  81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
  82
  83 static __read_mostly struct preempt_ops kvm_preempt_ops;
  84
  85 struct dentry *kvm_debugfs_dir;
  86
  87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
  88                            unsigned long arg);
  89 #ifdef CONFIG_COMPAT
  90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
  91                                   unsigned long arg);
  92 #endif
  93 static int hardware_enable_all(void);
  94 static void hardware_disable_all(void);
  95
  96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
  97
  98 bool kvm_rebooting;
  99 EXPORT_SYMBOL_GPL(kvm_rebooting);
 100
 101 static bool largepages_enabled = true;
 102
 103 bool kvm_is_mmio_pfn(pfn_t pfn)
 104 {
 105         if (pfn_valid(pfn)) {
 106                 int reserved;
 107                 struct page *tail = pfn_to_page(pfn);
 108                 struct page *head = compound_trans_head(tail);
 109                 reserved = PageReserved(head);
 110                 if (head != tail) {
 111                         /*
 112                          * "head" is not a dangling pointer
 113                          * (compound_trans_head takes care of that)
 114                          * but the hugepage may have been splitted
 115                          * from under us (and we may not hold a
 116                          * reference count on the head page so it can
 117                          * be reused before we run PageReferenced), so
 118                          * we've to check PageTail before returning
 119                          * what we just read.
 120                          */
 121                         smp_rmb();
 122                         if (PageTail(tail))
 123                                 return reserved;
 124                 }
 125                 return PageReserved(tail);
 126         }
 127
 128         return true;
 129 }
 130
 131 /*
 132  * Switches to specified vcpu, until a matching vcpu_put()
 133  */
 134 void vcpu_load(struct kvm_vcpu *vcpu)
 135 {
 136         int cpu;
 137
 138         mutex_lock(&vcpu->mutex);
 139         if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
 140                 /* The thread running this VCPU changed. */
 141                 struct pid *oldpid = vcpu->pid;
 142                 struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
 143                 rcu_assign_pointer(vcpu->pid, newpid);
 144                 synchronize_rcu();
 145                 put_pid(oldpid);
 146         }
 147         cpu = get_cpu();
 148         preempt_notifier_register(&vcpu->preempt_notifier);
 149         kvm_arch_vcpu_load(vcpu, cpu);
 150         put_cpu();
 151 }
 152
 153 void vcpu_put(struct kvm_vcpu *vcpu)
 154 {
 155         preempt_disable();
 156         kvm_arch_vcpu_put(vcpu);
 157         preempt_notifier_unregister(&vcpu->preempt_notifier);
 158         preempt_enable();
 159         mutex_unlock(&vcpu->mutex);
 160 }
 161
 162 static void ack_flush(void *_completed)
 163 {
 164 }
 165
 166 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
 167 {
 168         int i, cpu, me;
 169         cpumask_var_t cpus;
 170         bool called = true;
 171         struct kvm_vcpu *vcpu;
 172
 173         zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 174
 175         me = get_cpu();
 176         kvm_for_each_vcpu(i, vcpu, kvm) {
 177                 kvm_make_request(req, vcpu);
 178                 cpu = vcpu->cpu;
 179
 180                 /* Set ->requests bit before we read ->mode */
 181                 smp_mb();
 182
 183                 if (cpus != NULL && cpu != -1 && cpu != me &&
 184                       kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
 185                         cpumask_set_cpu(cpu, cpus);
 186         }
 187         if (unlikely(cpus == NULL))
 188                 smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
 189         else if (!cpumask_empty(cpus))
 190                 smp_call_function_many(cpus, ack_flush, NULL, 1);
 191         else
 192                 called = false;
 193         put_cpu();
 194         free_cpumask_var(cpus);
 195         return called;
 196 }
 197
 198 void kvm_flush_remote_tlbs(struct kvm *kvm)
 199 {
 200         long dirty_count = kvm->tlbs_dirty;
 201
 202         smp_mb();
 203         if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 204                 ++kvm->stat.remote_tlb_flush;
 205         cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 206 }
 207
 208 void kvm_reload_remote_mmus(struct kvm *kvm)
 209 {
 210         make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 211 }
 212
 213 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 214 {
 215         struct page *page;
 216         int r;
 217
 218         mutex_init(&vcpu->mutex);
 219         vcpu->cpu = -1;
 220         vcpu->kvm = kvm;
 221         vcpu->vcpu_id = id;
 222         vcpu->pid = NULL;
 223         init_waitqueue_head(&vcpu->wq);
 224         kvm_async_pf_vcpu_init(vcpu);
 225
 226         page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 227         if (!page) {
 228                 r = -ENOMEM;
 229                 goto fail;
 230         }
 231         vcpu->run = page_address(page);
 232
 233         kvm_vcpu_set_in_spin_loop(vcpu, false);
 234         kvm_vcpu_set_dy_eligible(vcpu, false);
 235
 236         r = kvm_arch_vcpu_init(vcpu);
 237         if (r < 0)
 238                 goto fail_free_run;
 239         return 0;
 240
 241 fail_free_run:
 242         free_page((unsigned long)vcpu->run);
 243 fail:
 244         return r;
 245 }
 246 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 247
 248 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 249 {
 250         put_pid(vcpu->pid);
 251         kvm_arch_vcpu_uninit(vcpu);
 252         free_page((unsigned long)vcpu->run);
 253 }
 254 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 255
 256 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 257 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 258 {
 259         return container_of(mn, struct kvm, mmu_notifier);
 260 }
 261
 262 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
 263                                              struct mm_struct *mm,
 264                                              unsigned long address)
 265 {
 266         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 267         int need_tlb_flush, idx;
 268
 269         /*
 270          * When ->invalidate_page runs, the linux pte has been zapped
 271          * already but the page is still allocated until
 272          * ->invalidate_page returns. So if we increase the sequence
 273          * here the kvm page fault will notice if the spte can't be
 274          * established because the page is going to be freed. If
 275          * instead the kvm page fault establishes the spte before
 276          * ->invalidate_page runs, kvm_unmap_hva will release it
 277          * before returning.
 278          *
 279          * The sequence increase only need to be seen at spin_unlock
 280          * time, and not at spin_lock time.
 281          *
 282          * Increasing the sequence after the spin_unlock would be
 283          * unsafe because the kvm page fault could then establish the
 284          * pte after kvm_unmap_hva returned, without noticing the page
 285          * is going to be freed.
 286          */
 287         idx = srcu_read_lock(&kvm->srcu);
 288         spin_lock(&kvm->mmu_lock);
 289
 290         kvm->mmu_notifier_seq++;
 291         need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
 292         /* we've to flush the tlb before the pages can be freed */
 293         if (need_tlb_flush)
 294                 kvm_flush_remote_tlbs(kvm);
 295
 296         spin_unlock(&kvm->mmu_lock);
 297         srcu_read_unlock(&kvm->srcu, idx);
 298 }
 299
 300 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 301                                         struct mm_struct *mm,
 302                                         unsigned long address,
 303                                         pte_t pte)
 304 {
 305         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 306         int idx;
 307
 308         idx = srcu_read_lock(&kvm->srcu);
 309         spin_lock(&kvm->mmu_lock);
 310         kvm->mmu_notifier_seq++;
 311         kvm_set_spte_hva(kvm, address, pte);
 312         spin_unlock(&kvm->mmu_lock);
 313         srcu_read_unlock(&kvm->srcu, idx);
 314 }
 315
 316 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 317                                                     struct mm_struct *mm,
 318                                                     unsigned long start,
 319                                                     unsigned long end)
 320 {
 321         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 322         int need_tlb_flush = 0, idx;
 323
 324         idx = srcu_read_lock(&kvm->srcu);
 325         spin_lock(&kvm->mmu_lock);
 326         /*
 327          * The count increase must become visible at unlock time as no
 328          * spte can be established without taking the mmu_lock and
 329          * count is also read inside the mmu_lock critical section.
 330          */
 331         kvm->mmu_notifier_count++;
 332         need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
 333         need_tlb_flush |= kvm->tlbs_dirty;
 334         /* we've to flush the tlb before the pages can be freed */
 335         if (need_tlb_flush)
 336                 kvm_flush_remote_tlbs(kvm);
 337
 338         spin_unlock(&kvm->mmu_lock);
 339         srcu_read_unlock(&kvm->srcu, idx);
 340 }
 341
 342 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 343                                                   struct mm_struct *mm,
 344                                                   unsigned long start,
 345                                                   unsigned long end)
 346 {
 347         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 348
 349         spin_lock(&kvm->mmu_lock);
 350         /*
 351          * This sequence increase will notify the kvm page fault that
 352          * the page that is going to be mapped in the spte could have
 353          * been freed.
 354          */
 355         kvm->mmu_notifier_seq++;
 356         smp_wmb();
 357         /*
 358          * The above sequence increase must be visible before the
 359          * below count decrease, which is ensured by the smp_wmb above
 360          * in conjunction with the smp_rmb in mmu_notifier_retry().
 361          */
 362         kvm->mmu_notifier_count--;
 363         spin_unlock(&kvm->mmu_lock);
 364
 365         BUG_ON(kvm->mmu_notifier_count < 0);
 366 }
 367
 368 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 369                                               struct mm_struct *mm,
 370                                               unsigned long address)
 371 {
 372         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 373         int young, idx;
 374
 375         idx = srcu_read_lock(&kvm->srcu);
 376         spin_lock(&kvm->mmu_lock);
 377
 378         young = kvm_age_hva(kvm, address);
 379         if (young)
 380                 kvm_flush_remote_tlbs(kvm);
 381
 382         spin_unlock(&kvm->mmu_lock);
 383         srcu_read_unlock(&kvm->srcu, idx);
 384
 385         return young;
 386 }
 387
 388 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 389                                        struct mm_struct *mm,
 390                                        unsigned long address)
 391 {
 392         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 393         int young, idx;
 394
 395         idx = srcu_read_lock(&kvm->srcu);
 396         spin_lock(&kvm->mmu_lock);
 397         young = kvm_test_age_hva(kvm, address);
 398         spin_unlock(&kvm->mmu_lock);
 399         srcu_read_unlock(&kvm->srcu, idx);
 400
 401         return young;
 402 }
 403
 404 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 405                                      struct mm_struct *mm)
 406 {
 407         struct kvm *kvm = mmu_notifier_to_kvm(mn);
 408         int idx;
 409
 410         idx = srcu_read_lock(&kvm->srcu);
 411         kvm_arch_flush_shadow(kvm);
 412         srcu_read_unlock(&kvm->srcu, idx);
 413 }
 414
 415 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 416         .invalidate_page        = kvm_mmu_notifier_invalidate_page,
 417         .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 418         .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 419         .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 420         .test_young             = kvm_mmu_notifier_test_young,
 421         .change_pte             = kvm_mmu_notifier_change_pte,
 422         .release                = kvm_mmu_notifier_release,
 423 };
 424
 425 static int kvm_init_mmu_notifier(struct kvm *kvm)
 426 {
 427         kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 428         return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 429 }
 430
 431 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 432
 433 static int kvm_init_mmu_notifier(struct kvm *kvm)
 434 {
 435         return 0;
 436 }
 437
 438 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 439
 440 static void kvm_init_memslots_id(struct kvm *kvm)
 441 {
 442         int i;
 443         struct kvm_memslots *slots = kvm->memslots;
 444
 445         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 446                 slots->id_to_index[i] = slots->memslots[i].id = i;
 447 }
 448
 449 static struct kvm *kvm_create_vm(unsigned long type)
 450 {
 451         int r, i;
 452         struct kvm *kvm = kvm_arch_alloc_vm();
 453
 454         if (!kvm)
 455                 return ERR_PTR(-ENOMEM);
 456
 457         r = kvm_arch_init_vm(kvm, type);
 458         if (r)
 459                 goto out_err_nodisable;
 460
 461         r = hardware_enable_all();
 462         if (r)
 463                 goto out_err_nodisable;
 464
 465 #ifdef CONFIG_HAVE_KVM_IRQCHIP
 466         INIT_HLIST_HEAD(&kvm->mask_notifier_list);
 467         INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 468 #endif
 469
 470         r = -ENOMEM;
 471         kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
 472         if (!kvm->memslots)
 473                 goto out_err_nosrcu;
 474         kvm_init_memslots_id(kvm);
 475         if (init_srcu_struct(&kvm->srcu))
 476                 goto out_err_nosrcu;
 477         for (i = 0; i < KVM_NR_BUSES; i++) {
 478                 kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
 479                                         GFP_KERNEL);
 480                 if (!kvm->buses[i])
 481                         goto out_err;
 482         }
 483
 484         spin_lock_init(&kvm->mmu_lock);
 485         kvm->mm = current->mm;
 486         atomic_inc(&kvm->mm->mm_count);
 487         kvm_eventfd_init(kvm);
 488         mutex_init(&kvm->lock);
 489         mutex_init(&kvm->irq_lock);
 490         mutex_init(&kvm->slots_lock);
 491         atomic_set(&kvm->users_count, 1);
 492
 493         r = kvm_init_mmu_notifier(kvm);
 494         if (r)
 495                 goto out_err;
 496
 497         raw_spin_lock(&kvm_lock);
 498         list_add(&kvm->vm_list, &vm_list);
 499         raw_spin_unlock(&kvm_lock);
 500
 501         return kvm;
 502
 503 out_err:
 504         cleanup_srcu_struct(&kvm->srcu);
 505 out_err_nosrcu:
 506         hardware_disable_all();
 507 out_err_nodisable:
 508         for (i = 0; i < KVM_NR_BUSES; i++)
 509                 kfree(kvm->buses[i]);
 510         kfree(kvm->memslots);
 511         kvm_arch_free_vm(kvm);
 512         return ERR_PTR(r);
 513 }
 514
 515 /*
 516  * Avoid using vmalloc for a small buffer.
 517  * Should not be used when the size is statically known.
 518  */
 519 void *kvm_kvzalloc(unsigned long size)
 520 {
 521         if (size > PAGE_SIZE)
 522                 return vzalloc(size);
 523         else
 524                 return kzalloc(size, GFP_KERNEL);
 525 }
 526
 527 void kvm_kvfree(const void *addr)
 528 {
 529         if (is_vmalloc_addr(addr))
 530                 vfree(addr);
 531         else
 532                 kfree(addr);
 533 }
 534
 535 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 536 {
 537         if (!memslot->dirty_bitmap)
 538                 return;
 539
 540         kvm_kvfree(memslot->dirty_bitmap);
 541         memslot->dirty_bitmap = NULL;
 542 }
 543
 544 /*
 545  * Free any memory in @free but not in @dont.
 546  */
 547 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
 548                                   struct kvm_memory_slot *dont)
 549 {
 550         if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 551                 kvm_destroy_dirty_bitmap(free);
 552
 553         kvm_arch_free_memslot(free, dont);
 554
 555         free->npages = 0;
 556 }
 557
 558 void kvm_free_physmem(struct kvm *kvm)
 559 {
 560         struct kvm_memslots *slots = kvm->memslots;
 561         struct kvm_memory_slot *memslot;
 562
 563         kvm_for_each_memslot(memslot, slots)
 564                 kvm_free_physmem_slot(memslot, NULL);
 565
 566         kfree(kvm->memslots);
 567 }
 568
 569 static void kvm_destroy_vm(struct kvm *kvm)
 570 {
 571         int i;
 572         struct mm_struct *mm = kvm->mm;
 573
 574         kvm_arch_sync_events(kvm);
 575         raw_spin_lock(&kvm_lock);
 576         list_del(&kvm->vm_list);
 577         raw_spin_unlock(&kvm_lock);
 578         kvm_free_irq_routing(kvm);
 579         for (i = 0; i < KVM_NR_BUSES; i++)
 580                 kvm_io_bus_destroy(kvm->buses[i]);
 581         kvm_coalesced_mmio_free(kvm);
 582 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 583         mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 584 #else
 585         kvm_arch_flush_shadow(kvm);
 586 #endif
 587         kvm_arch_destroy_vm(kvm);
 588         kvm_free_physmem(kvm);
 589         cleanup_srcu_struct(&kvm->srcu);
 590         kvm_arch_free_vm(kvm);
 591         hardware_disable_all();
 592         mmdrop(mm);
 593 }
 594
 595 void kvm_get_kvm(struct kvm *kvm)
 596 {
 597         atomic_inc(&kvm->users_count);
 598 }
 599 EXPORT_SYMBOL_GPL(kvm_get_kvm);
 600
 601 void kvm_put_kvm(struct kvm *kvm)
 602 {
 603         if (atomic_dec_and_test(&kvm->users_count))
 604                 kvm_destroy_vm(kvm);
 605 }
 606 EXPORT_SYMBOL_GPL(kvm_put_kvm);
 607
 608
 609 static int kvm_vm_release(struct inode *inode, struct file *filp)
 610 {
 611         struct kvm *kvm = filp->private_data;
 612
 613         kvm_irqfd_release(kvm);
 614
 615         kvm_put_kvm(kvm);
 616         return 0;
 617 }
 618
 619 /*
 620  * Allocation size is twice as large as the actual dirty bitmap size.
 621  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 622  */
 623 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 624 {
 625 #ifndef CONFIG_S390
 626         unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 627
 628         memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
 629         if (!memslot->dirty_bitmap)
 630                 return -ENOMEM;
 631
 632 #endif /* !CONFIG_S390 */
 633         return 0;
 634 }
 635
 636 static int cmp_memslot(const void *slot1, const void *slot2)
 637 {
 638         struct kvm_memory_slot *s1, *s2;
 639
 640         s1 = (struct kvm_memory_slot *)slot1;
 641         s2 = (struct kvm_memory_slot *)slot2;
 642
 643         if (s1->npages < s2->npages)
 644                 return 1;
 645         if (s1->npages > s2->npages)
 646                 return -1;
 647
 648         return 0;
 649 }
 650
 651 /*
 652  * Sort the memslots base on its size, so the larger slots
 653  * will get better fit.
 654  */
 655 static void sort_memslots(struct kvm_memslots *slots)
 656 {
 657         int i;
 658
 659         sort(slots->memslots, KVM_MEM_SLOTS_NUM,
 660               sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
 661
 662         for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 663                 slots->id_to_index[slots->memslots[i].id] = i;
 664 }
 665
 666 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
 667 {
 668         if (new) {
 669                 int id = new->id;
 670                 struct kvm_memory_slot *old = id_to_memslot(slots, id);
 671                 unsigned long npages = old->npages;
 672
 673                 *old = *new;
 674                 if (new->npages != npages)
 675                         sort_memslots(slots);
 676         }
 677
 678         slots->generation++;
 679 }
 680
 681 /*
 682  * Allocate some memory and give it an address in the guest physical address
 683  * space.
 684  *
 685  * Discontiguous memory is allowed, mostly for framebuffers.
 686  *
 687  * Must be called holding mmap_sem for write.
 688  */
 689 int __kvm_set_memory_region(struct kvm *kvm,
 690                             struct kvm_userspace_memory_region *mem,
 691                             int user_alloc)
 692 {
 693         int r;
 694         gfn_t base_gfn;
 695         unsigned long npages;
 696         unsigned long i;
 697         struct kvm_memory_slot *memslot;
 698         struct kvm_memory_slot old, new;
 699         struct kvm_memslots *slots, *old_memslots;
 700
 701         r = -EINVAL;
 702         /* General sanity checks */
 703         if (mem->memory_size & (PAGE_SIZE - 1))
 704                 goto out;
 705         if (mem->guest_phys_addr & (PAGE_SIZE - 1))
 706                 goto out;
 707         /* We can read the guest memory with __xxx_user() later on. */
 708         if (user_alloc &&
 709             ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
 710              !access_ok(VERIFY_WRITE,
 711                         (void __user *)(unsigned long)mem->userspace_addr,
 712                         mem->memory_size)))
 713                 goto out;
 714         if (mem->slot >= KVM_MEM_SLOTS_NUM)
 715                 goto out;
 716         if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
 717                 goto out;
 718
 719         memslot = id_to_memslot(kvm->memslots, mem->slot);
 720         base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
 721         npages = mem->memory_size >> PAGE_SHIFT;
 722
 723         r = -EINVAL;
 724         if (npages > KVM_MEM_MAX_NR_PAGES)
 725                 goto out;
 726
 727         if (!npages)
 728                 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
 729
 730         new = old = *memslot;
 731
 732         new.id = mem->slot;
 733         new.base_gfn = base_gfn;
 734         new.npages = npages;
 735         new.flags = mem->flags;
 736
 737         /* Disallow changing a memory slot's size. */
 738         r = -EINVAL;
 739         if (npages && old.npages && npages != old.npages)
 740                 goto out_free;
 741
 742         /* Check for overlaps */
 743         r = -EEXIST;
 744         for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
 745                 struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
 746
 747                 if (s == memslot || !s->npages)
 748                         continue;
 749                 if (!((base_gfn + npages <= s->base_gfn) ||
 750                       (base_gfn >= s->base_gfn + s->npages)))
 751                         goto out_free;
 752         }
 753
 754         /* Free page dirty bitmap if unneeded */
 755         if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
 756                 new.dirty_bitmap = NULL;
 757
 758         r = -ENOMEM;
 759
 760         /* Allocate if a slot is being created */
 761         if (npages && !old.npages) {
 762                 new.user_alloc = user_alloc;
 763                 new.userspace_addr = mem->userspace_addr;
 764
 765                 if (kvm_arch_create_memslot(&new, npages))
 766                         goto out_free;
 767         }
 768
 769         /* Allocate page dirty bitmap if needed */
 770         if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
 771                 if (kvm_create_dirty_bitmap(&new) < 0)
 772                         goto out_free;
 773                 /* destroy any largepage mappings for dirty tracking */
 774         }
 775
 776         if (!npages) {
 777                 struct kvm_memory_slot *slot;
 778
 779                 r = -ENOMEM;
 780                 slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 781                                 GFP_KERNEL);
 782                 if (!slots)
 783                         goto out_free;
 784                 slot = id_to_memslot(slots, mem->slot);
 785                 slot->flags |= KVM_MEMSLOT_INVALID;
 786
 787                 update_memslots(slots, NULL);
 788
 789                 old_memslots = kvm->memslots;
 790                 rcu_assign_pointer(kvm->memslots, slots);
 791                 synchronize_srcu_expedited(&kvm->srcu);
 792                 /* From this point no new shadow pages pointing to a deleted
 793                  * memslot will be created.
 794                  *
 795                  * validation of sp->gfn happens in:
 796                  *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
 797                  *      - kvm_is_visible_gfn (mmu_check_roots)
 798                  */
 799                 kvm_arch_flush_shadow(kvm);
 800                 kfree(old_memslots);
 801         }
 802
 803         r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
 804         if (r)
 805                 goto out_free;
 806
 807         /* map/unmap the pages in iommu page table */
 808         if (npages) {
 809                 r = kvm_iommu_map_pages(kvm, &new);
 810                 if (r)
 811                         goto out_free;
 812         } else
 813                 kvm_iommu_unmap_pages(kvm, &old);
 814
 815         r = -ENOMEM;
 816         slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
 817                         GFP_KERNEL);
 818         if (!slots)
 819                 goto out_free;
 820
 821         /* actual memory is freed via old in kvm_free_physmem_slot below */
 822         if (!npages) {
 823                 new.dirty_bitmap = NULL;
 824                 memset(&new.arch, 0, sizeof(new.arch));
 825         }
 826
 827         update_memslots(slots, &new);
 828         old_memslots = kvm->memslots;
 829         rcu_assign_pointer(kvm->memslots, slots);
 830         synchronize_srcu_expedited(&kvm->srcu);
 831
 832         kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
 833
 834         /*
 835          * If the new memory slot is created, we need to clear all
 836          * mmio sptes.
 837          */
 838         if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
 839                 kvm_arch_flush_shadow(kvm);
 840
 841         kvm_free_physmem_slot(&old, &new);
 842         kfree(old_memslots);
 843
 844         return 0;
 845
 846 out_free:
 847         kvm_free_physmem_slot(&new, &old);
 848 out:
 849         return r;
 850
 851 }
 852 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
 853
 854 int kvm_set_memory_region(struct kvm *kvm,
 855                           struct kvm_userspace_memory_region *mem,
 856                           int user_alloc)
 857 {
 858         int r;
 859
 860         mutex_lock(&kvm->slots_lock);
 861         r = __kvm_set_memory_region(kvm, mem, user_alloc);
 862         mutex_unlock(&kvm->slots_lock);
 863         return r;
 864 }
 865 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
 866
 867 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
 868                                    struct
 869                                    kvm_userspace_memory_region *mem,
 870                                    int user_alloc)
 871 {
 872         if (mem->slot >= KVM_MEMORY_SLOTS)
 873                 return -EINVAL;
 874         return kvm_set_memory_region(kvm, mem, user_alloc);
 875 }
 876
 877 int kvm_get_dirty_log(struct kvm *kvm,
 878                         struct kvm_dirty_log *log, int *is_dirty)
 879 {
 880         struct kvm_memory_slot *memslot;
 881         int r, i;
 882         unsigned long n;
 883         unsigned long any = 0;
 884
 885         r = -EINVAL;
 886         if (log->slot >= KVM_MEMORY_SLOTS)
 887                 goto out;
 888
 889         memslot = id_to_memslot(kvm->memslots, log->slot);
 890         r = -ENOENT;
 891         if (!memslot->dirty_bitmap)
 892                 goto out;
 893
 894         n = kvm_dirty_bitmap_bytes(memslot);
 895
 896         for (i = 0; !any && i < n/sizeof(long); ++i)
 897                 any = memslot->dirty_bitmap[i];
 898
 899         r = -EFAULT;
 900         if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
 901                 goto out;
 902
 903         if (any)
 904                 *is_dirty = 1;
 905
 906         r = 0;
 907 out:
 908         return r;
 909 }
 910
 911 bool kvm_largepages_enabled(void)
 912 {
 913         return largepages_enabled;
 914 }
 915
 916 void kvm_disable_largepages(void)
 917 {
 918         largepages_enabled = false;
 919 }
 920 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
 921
 922 static inline unsigned long bad_hva(void)
 923 {
 924         return PAGE_OFFSET;
 925 }
 926
 927 int kvm_is_error_hva(unsigned long addr)
 928 {
 929         return addr == bad_hva();
 930 }
 931 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
 932
 933 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
 934 {
 935         return __gfn_to_memslot(kvm_memslots(kvm), gfn);
 936 }
 937 EXPORT_SYMBOL_GPL(gfn_to_memslot);
 938
 939 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
 940 {
 941         struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
 942
 943         if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
 944               memslot->flags & KVM_MEMSLOT_INVALID)
 945                 return 0;
 946
 947         return 1;
 948 }
 949 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
 950
 951 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
 952 {
 953         struct vm_area_struct *vma;
 954         unsigned long addr, size;
 955
 956         size = PAGE_SIZE;
 957
 958         addr = gfn_to_hva(kvm, gfn);
 959         if (kvm_is_error_hva(addr))
 960                 return PAGE_SIZE;
 961
 962         down_read(&current->mm->mmap_sem);
 963         vma = find_vma(current->mm, addr);
 964         if (!vma)
 965                 goto out;
 966
 967         size = vma_kernel_pagesize(vma);
 968
 969 out:
 970         up_read(&current->mm->mmap_sem);
 971
 972         return size;
 973 }
 974
 975 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
 976                                      gfn_t *nr_pages)
 977 {
 978         if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
 979                 return bad_hva();
 980
 981         if (nr_pages)
 982                 *nr_pages = slot->npages - (gfn - slot->base_gfn);
 983
 984         return gfn_to_hva_memslot(slot, gfn);
 985 }
 986
 987 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
 988 {
 989         return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
 990 }
 991 EXPORT_SYMBOL_GPL(gfn_to_hva);
 992
 993 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
 994         unsigned long start, int write, struct page **page)
 995 {
 996         int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
 997
 998         if (write)
 999                 flags |= FOLL_WRITE;
1000
1001         return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1002 }
1003
1004 static inline int check_user_page_hwpoison(unsigned long addr)
1005 {
1006         int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1007
1008         rc = __get_user_pages(current, current->mm, addr, 1,
1009                               flags, NULL, NULL, NULL);
1010         return rc == -EHWPOISON;
1011 }
1012
1013 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1014                         bool write_fault, bool *writable)
1015 {
1016         struct page *page[1];
1017         int npages = 0;
1018         pfn_t pfn;
1019
1020         /* we can do it either atomically or asynchronously, not both */
1021         BUG_ON(atomic && async);
1022
1023         BUG_ON(!write_fault && !writable);
1024
1025         if (writable)
1026                 *writable = true;
1027
1028         if (atomic || async)
1029                 npages = __get_user_pages_fast(addr, 1, 1, page);
1030
1031         if (unlikely(npages != 1) && !atomic) {
1032                 might_sleep();
1033
1034                 if (writable)
1035                         *writable = write_fault;
1036
1037                 if (async) {
1038                         down_read(&current->mm->mmap_sem);
1039                         npages = get_user_page_nowait(current, current->mm,
1040                                                      addr, write_fault, page);
1041                         up_read(&current->mm->mmap_sem);
1042                 } else
1043                         npages = get_user_pages_fast(addr, 1, write_fault,
1044                                                      page);
1045
1046                 /* map read fault as writable if possible */
1047                 if (unlikely(!write_fault) && npages == 1) {
1048                         struct page *wpage[1];
1049
1050                         npages = __get_user_pages_fast(addr, 1, 1, wpage);
1051                         if (npages == 1) {
1052                                 *writable = true;
1053                                 put_page(page[0]);
1054                                 page[0] = wpage[0];
1055                         }
1056                         npages = 1;
1057                 }
1058         }
1059
1060         if (unlikely(npages != 1)) {
1061                 struct vm_area_struct *vma;
1062
1063                 if (atomic)
1064                         return KVM_PFN_ERR_FAULT;
1065
1066                 down_read(&current->mm->mmap_sem);
1067                 if (npages == -EHWPOISON ||
1068                         (!async && check_user_page_hwpoison(addr))) {
1069                         up_read(&current->mm->mmap_sem);
1070                         return KVM_PFN_ERR_HWPOISON;
1071                 }
1072
1073                 vma = find_vma_intersection(current->mm, addr, addr+1);
1074
1075                 if (vma == NULL)
1076                         pfn = KVM_PFN_ERR_FAULT;
1077                 else if ((vma->vm_flags & VM_PFNMAP)) {
1078                         pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1079                                 vma->vm_pgoff;
1080                         BUG_ON(!kvm_is_mmio_pfn(pfn));
1081                 } else {
1082                         if (async && (vma->vm_flags & VM_WRITE))
1083                                 *async = true;
1084                         pfn = KVM_PFN_ERR_FAULT;
1085                 }
1086                 up_read(&current->mm->mmap_sem);
1087         } else
1088                 pfn = page_to_pfn(page[0]);
1089
1090         return pfn;
1091 }
1092
1093 pfn_t hva_to_pfn_atomic(unsigned long addr)
1094 {
1095         return hva_to_pfn(addr, true, NULL, true, NULL);
1096 }
1097 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1098
1099 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1100                           bool write_fault, bool *writable)
1101 {
1102         unsigned long addr;
1103
1104         if (async)
1105                 *async = false;
1106
1107         addr = gfn_to_hva(kvm, gfn);
1108         if (kvm_is_error_hva(addr))
1109                 return KVM_PFN_ERR_BAD;
1110
1111         return hva_to_pfn(addr, atomic, async, write_fault, writable);
1112 }
1113
1114 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1115 {
1116         return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1117 }
1118 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1119
1120 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1121                        bool write_fault, bool *writable)
1122 {
1123         return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1124 }
1125 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1126
1127 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1128 {
1129         return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1130 }
1131 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1132
1133 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1134                       bool *writable)
1135 {
1136         return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1137 }
1138 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1139
1140 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1141 {
1142         unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1143         return hva_to_pfn(addr, false, NULL, true, NULL);
1144 }
1145
1146 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1147                                                                   int nr_pages)
1148 {
1149         unsigned long addr;
1150         gfn_t entry;
1151
1152         addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1153         if (kvm_is_error_hva(addr))
1154                 return -1;
1155
1156         if (entry < nr_pages)
1157                 return 0;
1158
1159         return __get_user_pages_fast(addr, nr_pages, 1, pages);
1160 }
1161 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1162
1163 static struct page *kvm_pfn_to_page(pfn_t pfn)
1164 {
1165         if (is_error_pfn(pfn))
1166                 return KVM_ERR_PTR_BAD_PAGE;
1167
1168         if (kvm_is_mmio_pfn(pfn)) {
1169                 WARN_ON(1);
1170                 return KVM_ERR_PTR_BAD_PAGE;
1171         }
1172
1173         return pfn_to_page(pfn);
1174 }
1175
1176 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1177 {
1178         pfn_t pfn;
1179
1180         pfn = gfn_to_pfn(kvm, gfn);
1181
1182         return kvm_pfn_to_page(pfn);
1183 }
1184
1185 EXPORT_SYMBOL_GPL(gfn_to_page);
1186
1187 void kvm_release_page_clean(struct page *page)
1188 {
1189         WARN_ON(is_error_page(page));
1190
1191         kvm_release_pfn_clean(page_to_pfn(page));
1192 }
1193 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1194
1195 void kvm_release_pfn_clean(pfn_t pfn)
1196 {
1197         WARN_ON(is_error_pfn(pfn));
1198
1199         if (!kvm_is_mmio_pfn(pfn))
1200                 put_page(pfn_to_page(pfn));
1201 }
1202 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1203
1204 void kvm_release_page_dirty(struct page *page)
1205 {
1206         WARN_ON(is_error_page(page));
1207
1208         kvm_release_pfn_dirty(page_to_pfn(page));
1209 }
1210 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1211
1212 void kvm_release_pfn_dirty(pfn_t pfn)
1213 {
1214         kvm_set_pfn_dirty(pfn);
1215         kvm_release_pfn_clean(pfn);
1216 }
1217 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1218
1219 void kvm_set_page_dirty(struct page *page)
1220 {
1221         kvm_set_pfn_dirty(page_to_pfn(page));
1222 }
1223 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1224
1225 void kvm_set_pfn_dirty(pfn_t pfn)
1226 {
1227         if (!kvm_is_mmio_pfn(pfn)) {
1228                 struct page *page = pfn_to_page(pfn);
1229                 if (!PageReserved(page))
1230                         SetPageDirty(page);
1231         }
1232 }
1233 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1234
1235 void kvm_set_pfn_accessed(pfn_t pfn)
1236 {
1237         if (!kvm_is_mmio_pfn(pfn))
1238                 mark_page_accessed(pfn_to_page(pfn));
1239 }
1240 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1241
1242 void kvm_get_pfn(pfn_t pfn)
1243 {
1244         if (!kvm_is_mmio_pfn(pfn))
1245                 get_page(pfn_to_page(pfn));
1246 }
1247 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1248
1249 static int next_segment(unsigned long len, int offset)
1250 {
1251         if (len > PAGE_SIZE - offset)
1252                 return PAGE_SIZE - offset;
1253         else
1254                 return len;
1255 }
1256
1257 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1258                         int len)
1259 {
1260         int r;
1261         unsigned long addr;
1262
1263         addr = gfn_to_hva(kvm, gfn);
1264         if (kvm_is_error_hva(addr))
1265                 return -EFAULT;
1266         r = __copy_from_user(data, (void __user *)addr + offset, len);
1267         if (r)
1268                 return -EFAULT;
1269         return 0;
1270 }
1271 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1272
1273 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1274 {
1275         gfn_t gfn = gpa >> PAGE_SHIFT;
1276         int seg;
1277         int offset = offset_in_page(gpa);
1278         int ret;
1279
1280         while ((seg = next_segment(len, offset)) != 0) {
1281                 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1282                 if (ret < 0)
1283                         return ret;
1284                 offset = 0;
1285                 len -= seg;
1286                 data += seg;
1287                 ++gfn;
1288         }
1289         return 0;
1290 }
1291 EXPORT_SYMBOL_GPL(kvm_read_guest);
1292
1293 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1294                           unsigned long len)
1295 {
1296         int r;
1297         unsigned long addr;
1298         gfn_t gfn = gpa >> PAGE_SHIFT;
1299         int offset = offset_in_page(gpa);
1300
1301         addr = gfn_to_hva(kvm, gfn);
1302         if (kvm_is_error_hva(addr))
1303                 return -EFAULT;
1304         pagefault_disable();
1305         r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1306         pagefault_enable();
1307         if (r)
1308                 return -EFAULT;
1309         return 0;
1310 }
1311 EXPORT_SYMBOL(kvm_read_guest_atomic);
1312
1313 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1314                          int offset, int len)
1315 {
1316         int r;
1317         unsigned long addr;
1318
1319         addr = gfn_to_hva(kvm, gfn);
1320         if (kvm_is_error_hva(addr))
1321                 return -EFAULT;
1322         r = __copy_to_user((void __user *)addr + offset, data, len);
1323         if (r)
1324                 return -EFAULT;
1325         mark_page_dirty(kvm, gfn);
1326         return 0;
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1329
1330 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1331                     unsigned long len)
1332 {
1333         gfn_t gfn = gpa >> PAGE_SHIFT;
1334         int seg;
1335         int offset = offset_in_page(gpa);
1336         int ret;
1337
1338         while ((seg = next_segment(len, offset)) != 0) {
1339                 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1340                 if (ret < 0)
1341                         return ret;
1342                 offset = 0;
1343                 len -= seg;
1344                 data += seg;
1345                 ++gfn;
1346         }
1347         return 0;
1348 }
1349
1350 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1351                               gpa_t gpa)
1352 {
1353         struct kvm_memslots *slots = kvm_memslots(kvm);
1354         int offset = offset_in_page(gpa);
1355         gfn_t gfn = gpa >> PAGE_SHIFT;
1356
1357         ghc->gpa = gpa;
1358         ghc->generation = slots->generation;
1359         ghc->memslot = gfn_to_memslot(kvm, gfn);
1360         ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1361         if (!kvm_is_error_hva(ghc->hva))
1362                 ghc->hva += offset;
1363         else
1364                 return -EFAULT;
1365
1366         return 0;
1367 }
1368 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1369
1370 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1371                            void *data, unsigned long len)
1372 {
1373         struct kvm_memslots *slots = kvm_memslots(kvm);
1374         int r;
1375
1376         if (slots->generation != ghc->generation)
1377                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1378
1379         if (kvm_is_error_hva(ghc->hva))
1380                 return -EFAULT;
1381
1382         r = __copy_to_user((void __user *)ghc->hva, data, len);
1383         if (r)
1384                 return -EFAULT;
1385         mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1386
1387         return 0;
1388 }
1389 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1390
1391 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1392                            void *data, unsigned long len)
1393 {
1394         struct kvm_memslots *slots = kvm_memslots(kvm);
1395         int r;
1396
1397         if (slots->generation != ghc->generation)
1398                 kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1399
1400         if (kvm_is_error_hva(ghc->hva))
1401                 return -EFAULT;
1402
1403         r = __copy_from_user(data, (void __user *)ghc->hva, len);
1404         if (r)
1405                 return -EFAULT;
1406
1407         return 0;
1408 }
1409 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1410
1411 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1412 {
1413         return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1414                                     offset, len);
1415 }
1416 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1417
1418 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1419 {
1420         gfn_t gfn = gpa >> PAGE_SHIFT;
1421         int seg;
1422         int offset = offset_in_page(gpa);
1423         int ret;
1424
1425         while ((seg = next_segment(len, offset)) != 0) {
1426                 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1427                 if (ret < 0)
1428                         return ret;
1429                 offset = 0;
1430                 len -= seg;
1431                 ++gfn;
1432         }
1433         return 0;
1434 }
1435 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1436
1437 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1438                              gfn_t gfn)
1439 {
1440         if (memslot && memslot->dirty_bitmap) {
1441                 unsigned long rel_gfn = gfn - memslot->base_gfn;
1442
1443                 /* TODO: introduce set_bit_le() and use it */
1444                 test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap);
1445         }
1446 }
1447
1448 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1449 {
1450         struct kvm_memory_slot *memslot;
1451
1452         memslot = gfn_to_memslot(kvm, gfn);
1453         mark_page_dirty_in_slot(kvm, memslot, gfn);
1454 }
1455
1456 /*
1457  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1458  */
1459 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1460 {
1461         DEFINE_WAIT(wait);
1462
1463         for (;;) {
1464                 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1465
1466                 if (kvm_arch_vcpu_runnable(vcpu)) {
1467                         kvm_make_request(KVM_REQ_UNHALT, vcpu);
1468                         break;
1469                 }
1470                 if (kvm_cpu_has_pending_timer(vcpu))
1471                         break;
1472                 if (signal_pending(current))
1473                         break;
1474
1475                 schedule();
1476         }
1477
1478         finish_wait(&vcpu->wq, &wait);
1479 }
1480
1481 #ifndef CONFIG_S390
1482 /*
1483  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
1484  */
1485 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
1486 {
1487         int me;
1488         int cpu = vcpu->cpu;
1489         wait_queue_head_t *wqp;
1490
1491         wqp = kvm_arch_vcpu_wq(vcpu);
1492         if (waitqueue_active(wqp)) {
1493                 wake_up_interruptible(wqp);
1494                 ++vcpu->stat.halt_wakeup;
1495         }
1496
1497         me = get_cpu();
1498         if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
1499                 if (kvm_arch_vcpu_should_kick(vcpu))
1500                         smp_send_reschedule(cpu);
1501         put_cpu();
1502 }
1503 #endif /* !CONFIG_S390 */
1504
1505 void kvm_resched(struct kvm_vcpu *vcpu)
1506 {
1507         if (!need_resched())
1508                 return;
1509         cond_resched();
1510 }
1511 EXPORT_SYMBOL_GPL(kvm_resched);
1512
1513 bool kvm_vcpu_yield_to(struct kvm_vcpu *target)
1514 {
1515         struct pid *pid;
1516         struct task_struct *task = NULL;
1517
1518         rcu_read_lock();
1519         pid = rcu_dereference(target->pid);
1520         if (pid)
1521                 task = get_pid_task(target->pid, PIDTYPE_PID);
1522         rcu_read_unlock();
1523         if (!task)
1524                 return false;
1525         if (task->flags & PF_VCPU) {
1526                 put_task_struct(task);
1527                 return false;
1528         }
1529         if (yield_to(task, 1)) {
1530                 put_task_struct(task);
1531                 return true;
1532         }
1533         put_task_struct(task);
1534         return false;
1535 }
1536 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
1537
1538 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
1539 /*
1540  * Helper that checks whether a VCPU is eligible for directed yield.
1541  * Most eligible candidate to yield is decided by following heuristics:
1542  *
1543  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
1544  *  (preempted lock holder), indicated by @in_spin_loop.
1545  *  Set at the beiginning and cleared at the end of interception/PLE handler.
1546  *
1547  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
1548  *  chance last time (mostly it has become eligible now since we have probably
1549  *  yielded to lockholder in last iteration. This is done by toggling
1550  *  @dy_eligible each time a VCPU checked for eligibility.)
1551  *
1552  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
1553  *  to preempted lock-holder could result in wrong VCPU selection and CPU
1554  *  burning. Giving priority for a potential lock-holder increases lock
1555  *  progress.
1556  *
1557  *  Since algorithm is based on heuristics, accessing another VCPU data without
1558  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
1559  *  and continue with next VCPU and so on.
1560  */
1561 bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
1562 {
1563         bool eligible;
1564
1565         eligible = !vcpu->spin_loop.in_spin_loop ||
1566                         (vcpu->spin_loop.in_spin_loop &&
1567                          vcpu->spin_loop.dy_eligible);
1568
1569         if (vcpu->spin_loop.in_spin_loop)
1570                 kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
1571
1572         return eligible;
1573 }
1574 #endif
1575 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1576 {
1577         struct kvm *kvm = me->kvm;
1578         struct kvm_vcpu *vcpu;
1579         int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1580         int yielded = 0;
1581         int pass;
1582         int i;
1583
1584         kvm_vcpu_set_in_spin_loop(me, true);
1585         /*
1586          * We boost the priority of a VCPU that is runnable but not
1587          * currently running, because it got preempted by something
1588          * else and called schedule in __vcpu_run.  Hopefully that
1589          * VCPU is holding the lock that we need and will release it.
1590          * We approximate round-robin by starting at the last boosted VCPU.
1591          */
1592         for (pass = 0; pass < 2 && !yielded; pass++) {
1593                 kvm_for_each_vcpu(i, vcpu, kvm) {
1594                         if (!pass && i <= last_boosted_vcpu) {
1595                                 i = last_boosted_vcpu;
1596                                 continue;
1597                         } else if (pass && i > last_boosted_vcpu)
1598                                 break;
1599                         if (vcpu == me)
1600                                 continue;
1601                         if (waitqueue_active(&vcpu->wq))
1602                                 continue;
1603                         if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
1604                                 continue;
1605                         if (kvm_vcpu_yield_to(vcpu)) {
1606                                 kvm->last_boosted_vcpu = i;
1607                                 yielded = 1;
1608                                 break;
1609                         }
1610                 }
1611         }
1612         kvm_vcpu_set_in_spin_loop(me, false);
1613
1614         /* Ensure vcpu is not eligible during next spinloop */
1615         kvm_vcpu_set_dy_eligible(me, false);
1616 }
1617 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1618
1619 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1620 {
1621         struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1622         struct page *page;
1623
1624         if (vmf->pgoff == 0)
1625                 page = virt_to_page(vcpu->run);
1626 #ifdef CONFIG_X86
1627         else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1628                 page = virt_to_page(vcpu->arch.pio_data);
1629 #endif
1630 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1631         else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1632                 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1633 #endif
1634         else
1635                 return kvm_arch_vcpu_fault(vcpu, vmf);
1636         get_page(page);
1637         vmf->page = page;
1638         return 0;
1639 }
1640
1641 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1642         .fault = kvm_vcpu_fault,
1643 };
1644
1645 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1646 {
1647         vma->vm_ops = &kvm_vcpu_vm_ops;
1648         return 0;
1649 }
1650
1651 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1652 {
1653         struct kvm_vcpu *vcpu = filp->private_data;
1654
1655         kvm_put_kvm(vcpu->kvm);
1656         return 0;
1657 }
1658
1659 static struct file_operations kvm_vcpu_fops = {
1660         .release        = kvm_vcpu_release,
1661         .unlocked_ioctl = kvm_vcpu_ioctl,
1662 #ifdef CONFIG_COMPAT
1663         .compat_ioctl   = kvm_vcpu_compat_ioctl,
1664 #endif
1665         .mmap           = kvm_vcpu_mmap,
1666         .llseek         = noop_llseek,
1667 };
1668
1669 /*
1670  * Allocates an inode for the vcpu.
1671  */
1672 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1673 {
1674         return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1675 }
1676
1677 /*
1678  * Creates some virtual cpus.  Good luck creating more than one.
1679  */
1680 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1681 {
1682         int r;
1683         struct kvm_vcpu *vcpu, *v;
1684
1685         vcpu = kvm_arch_vcpu_create(kvm, id);
1686         if (IS_ERR(vcpu))
1687                 return PTR_ERR(vcpu);
1688
1689         preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1690
1691         r = kvm_arch_vcpu_setup(vcpu);
1692         if (r)
1693                 goto vcpu_destroy;
1694
1695         mutex_lock(&kvm->lock);
1696         if (!kvm_vcpu_compatible(vcpu)) {
1697                 r = -EINVAL;
1698                 goto unlock_vcpu_destroy;
1699         }
1700         if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1701                 r = -EINVAL;
1702                 goto unlock_vcpu_destroy;
1703         }
1704
1705         kvm_for_each_vcpu(r, v, kvm)
1706                 if (v->vcpu_id == id) {
1707                         r = -EEXIST;
1708                         goto unlock_vcpu_destroy;
1709                 }
1710
1711         BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1712
1713         /* Now it's all set up, let userspace reach it */
1714         kvm_get_kvm(kvm);
1715         r = create_vcpu_fd(vcpu);
1716         if (r < 0) {
1717                 kvm_put_kvm(kvm);
1718                 goto unlock_vcpu_destroy;
1719         }
1720
1721         kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1722         smp_wmb();
1723         atomic_inc(&kvm->online_vcpus);
1724
1725         mutex_unlock(&kvm->lock);
1726         return r;
1727
1728 unlock_vcpu_destroy:
1729         mutex_unlock(&kvm->lock);
1730 vcpu_destroy:
1731         kvm_arch_vcpu_destroy(vcpu);
1732         return r;
1733 }
1734
1735 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1736 {
1737         if (sigset) {
1738                 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1739                 vcpu->sigset_active = 1;
1740                 vcpu->sigset = *sigset;
1741         } else
1742                 vcpu->sigset_active = 0;
1743         return 0;
1744 }
1745
1746 static long kvm_vcpu_ioctl(struct file *filp,
1747                            unsigned int ioctl, unsigned long arg)
1748 {
1749         struct kvm_vcpu *vcpu = filp->private_data;
1750         void __user *argp = (void __user *)arg;
1751         int r;
1752         struct kvm_fpu *fpu = NULL;
1753         struct kvm_sregs *kvm_sregs = NULL;
1754
1755         if (vcpu->kvm->mm != current->mm)
1756                 return -EIO;
1757
1758 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1759         /*
1760          * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1761          * so vcpu_load() would break it.
1762          */
1763         if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1764                 return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1765 #endif
1766
1767
1768         vcpu_load(vcpu);
1769         switch (ioctl) {
1770         case KVM_RUN:
1771                 r = -EINVAL;
1772                 if (arg)
1773                         goto out;
1774                 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1775                 trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1776                 break;
1777         case KVM_GET_REGS: {
1778                 struct kvm_regs *kvm_regs;
1779
1780                 r = -ENOMEM;
1781                 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1782                 if (!kvm_regs)
1783                         goto out;
1784                 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1785                 if (r)
1786                         goto out_free1;
1787                 r = -EFAULT;
1788                 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1789                         goto out_free1;
1790                 r = 0;
1791 out_free1:
1792                 kfree(kvm_regs);
1793                 break;
1794         }
1795         case KVM_SET_REGS: {
1796                 struct kvm_regs *kvm_regs;
1797
1798                 r = -ENOMEM;
1799                 kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1800                 if (IS_ERR(kvm_regs)) {
1801                         r = PTR_ERR(kvm_regs);
1802                         goto out;
1803                 }
1804                 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1805                 if (r)
1806                         goto out_free2;
1807                 r = 0;
1808 out_free2:
1809                 kfree(kvm_regs);
1810                 break;
1811         }
1812         case KVM_GET_SREGS: {
1813                 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1814                 r = -ENOMEM;
1815                 if (!kvm_sregs)
1816                         goto out;
1817                 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1818                 if (r)
1819                         goto out;
1820                 r = -EFAULT;
1821                 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1822                         goto out;
1823                 r = 0;
1824                 break;
1825         }
1826         case KVM_SET_SREGS: {
1827                 kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1828                 if (IS_ERR(kvm_sregs)) {
1829                         r = PTR_ERR(kvm_sregs);
1830                         goto out;
1831                 }
1832                 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1833                 if (r)
1834                         goto out;
1835                 r = 0;
1836                 break;
1837         }
1838         case KVM_GET_MP_STATE: {
1839                 struct kvm_mp_state mp_state;
1840
1841                 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1842                 if (r)
1843                         goto out;
1844                 r = -EFAULT;
1845                 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1846                         goto out;
1847                 r = 0;
1848                 break;
1849         }
1850         case KVM_SET_MP_STATE: {
1851                 struct kvm_mp_state mp_state;
1852
1853                 r = -EFAULT;
1854                 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1855                         goto out;
1856                 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1857                 if (r)
1858                         goto out;
1859                 r = 0;
1860                 break;
1861         }
1862         case KVM_TRANSLATE: {
1863                 struct kvm_translation tr;
1864
1865                 r = -EFAULT;
1866                 if (copy_from_user(&tr, argp, sizeof tr))
1867                         goto out;
1868                 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1869                 if (r)
1870                         goto out;
1871                 r = -EFAULT;
1872                 if (copy_to_user(argp, &tr, sizeof tr))
1873                         goto out;
1874                 r = 0;
1875                 break;
1876         }
1877         case KVM_SET_GUEST_DEBUG: {
1878                 struct kvm_guest_debug dbg;
1879
1880                 r = -EFAULT;
1881                 if (copy_from_user(&dbg, argp, sizeof dbg))
1882                         goto out;
1883                 r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1884                 if (r)
1885                         goto out;
1886                 r = 0;
1887                 break;
1888         }
1889         case KVM_SET_SIGNAL_MASK: {
1890                 struct kvm_signal_mask __user *sigmask_arg = argp;
1891                 struct kvm_signal_mask kvm_sigmask;
1892                 sigset_t sigset, *p;
1893
1894                 p = NULL;
1895                 if (argp) {
1896                         r = -EFAULT;
1897                         if (copy_from_user(&kvm_sigmask, argp,
1898                                            sizeof kvm_sigmask))
1899                                 goto out;
1900                         r = -EINVAL;
1901                         if (kvm_sigmask.len != sizeof sigset)
1902                                 goto out;
1903                         r = -EFAULT;
1904                         if (copy_from_user(&sigset, sigmask_arg->sigset,
1905                                            sizeof sigset))
1906                                 goto out;
1907                         p = &sigset;
1908                 }
1909                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1910                 break;
1911         }
1912         case KVM_GET_FPU: {
1913                 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1914                 r = -ENOMEM;
1915                 if (!fpu)
1916                         goto out;
1917                 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1918                 if (r)
1919                         goto out;
1920                 r = -EFAULT;
1921                 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1922                         goto out;
1923                 r = 0;
1924                 break;
1925         }
1926         case KVM_SET_FPU: {
1927                 fpu = memdup_user(argp, sizeof(*fpu));
1928                 if (IS_ERR(fpu)) {
1929                         r = PTR_ERR(fpu);
1930                         goto out;
1931                 }
1932                 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1933                 if (r)
1934                         goto out;
1935                 r = 0;
1936                 break;
1937         }
1938         default:
1939                 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1940         }
1941 out:
1942         vcpu_put(vcpu);
1943         kfree(fpu);
1944         kfree(kvm_sregs);
1945         return r;
1946 }
1947
1948 #ifdef CONFIG_COMPAT
1949 static long kvm_vcpu_compat_ioctl(struct file *filp,
1950                                   unsigned int ioctl, unsigned long arg)
1951 {
1952         struct kvm_vcpu *vcpu = filp->private_data;
1953         void __user *argp = compat_ptr(arg);
1954         int r;
1955
1956         if (vcpu->kvm->mm != current->mm)
1957                 return -EIO;
1958
1959         switch (ioctl) {
1960         case KVM_SET_SIGNAL_MASK: {
1961                 struct kvm_signal_mask __user *sigmask_arg = argp;
1962                 struct kvm_signal_mask kvm_sigmask;
1963                 compat_sigset_t csigset;
1964                 sigset_t sigset;
1965
1966                 if (argp) {
1967                         r = -EFAULT;
1968                         if (copy_from_user(&kvm_sigmask, argp,
1969                                            sizeof kvm_sigmask))
1970                                 goto out;
1971                         r = -EINVAL;
1972                         if (kvm_sigmask.len != sizeof csigset)
1973                                 goto out;
1974                         r = -EFAULT;
1975                         if (copy_from_user(&csigset, sigmask_arg->sigset,
1976                                            sizeof csigset))
1977                                 goto out;
1978                 }
1979                 sigset_from_compat(&sigset, &csigset);
1980                 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1981                 break;
1982         }
1983         default:
1984                 r = kvm_vcpu_ioctl(filp, ioctl, arg);
1985         }
1986
1987 out:
1988         return r;
1989 }
1990 #endif
1991
1992 static long kvm_vm_ioctl(struct file *filp,
1993                            unsigned int ioctl, unsigned long arg)
1994 {
1995         struct kvm *kvm = filp->private_data;
1996         void __user *argp = (void __user *)arg;
1997         int r;
1998
1999         if (kvm->mm != current->mm)
2000                 return -EIO;
2001         switch (ioctl) {
2002         case KVM_CREATE_VCPU:
2003                 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2004                 if (r < 0)
2005                         goto out;
2006                 break;
2007         case KVM_SET_USER_MEMORY_REGION: {
2008                 struct kvm_userspace_memory_region kvm_userspace_mem;
2009
2010                 r = -EFAULT;
2011                 if (copy_from_user(&kvm_userspace_mem, argp,
2012                                                 sizeof kvm_userspace_mem))
2013                         goto out;
2014
2015                 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
2016                 if (r)
2017                         goto out;
2018                 break;
2019         }
2020         case KVM_GET_DIRTY_LOG: {
2021                 struct kvm_dirty_log log;
2022
2023                 r = -EFAULT;
2024                 if (copy_from_user(&log, argp, sizeof log))
2025                         goto out;
2026                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2027                 if (r)
2028                         goto out;
2029                 break;
2030         }
2031 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2032         case KVM_REGISTER_COALESCED_MMIO: {
2033                 struct kvm_coalesced_mmio_zone zone;
2034                 r = -EFAULT;
2035                 if (copy_from_user(&zone, argp, sizeof zone))
2036                         goto out;
2037                 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2038                 if (r)
2039                         goto out;
2040                 r = 0;
2041                 break;
2042         }
2043         case KVM_UNREGISTER_COALESCED_MMIO: {
2044                 struct kvm_coalesced_mmio_zone zone;
2045                 r = -EFAULT;
2046                 if (copy_from_user(&zone, argp, sizeof zone))
2047                         goto out;
2048                 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2049                 if (r)
2050                         goto out;
2051                 r = 0;
2052                 break;
2053         }
2054 #endif
2055         case KVM_IRQFD: {
2056                 struct kvm_irqfd data;
2057
2058                 r = -EFAULT;
2059                 if (copy_from_user(&data, argp, sizeof data))
2060                         goto out;
2061                 r = kvm_irqfd(kvm, &data);
2062                 break;
2063         }
2064         case KVM_IOEVENTFD: {
2065                 struct kvm_ioeventfd data;
2066
2067                 r = -EFAULT;
2068                 if (copy_from_user(&data, argp, sizeof data))
2069                         goto out;
2070                 r = kvm_ioeventfd(kvm, &data);
2071                 break;
2072         }
2073 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2074         case KVM_SET_BOOT_CPU_ID:
2075                 r = 0;
2076                 mutex_lock(&kvm->lock);
2077                 if (atomic_read(&kvm->online_vcpus) != 0)
2078                         r = -EBUSY;
2079                 else
2080                         kvm->bsp_vcpu_id = arg;
2081                 mutex_unlock(&kvm->lock);
2082                 break;
2083 #endif
2084 #ifdef CONFIG_HAVE_KVM_MSI
2085         case KVM_SIGNAL_MSI: {
2086                 struct kvm_msi msi;
2087
2088                 r = -EFAULT;
2089                 if (copy_from_user(&msi, argp, sizeof msi))
2090                         goto out;
2091                 r = kvm_send_userspace_msi(kvm, &msi);
2092                 break;
2093         }
2094 #endif
2095 #ifdef __KVM_HAVE_IRQ_LINE
2096         case KVM_IRQ_LINE_STATUS:
2097         case KVM_IRQ_LINE: {
2098                 struct kvm_irq_level irq_event;
2099
2100                 r = -EFAULT;
2101                 if (copy_from_user(&irq_event, argp, sizeof irq_event))
2102                         goto out;
2103
2104                 r = kvm_vm_ioctl_irq_line(kvm, &irq_event);
2105                 if (r)
2106                         goto out;
2107
2108                 r = -EFAULT;
2109                 if (ioctl == KVM_IRQ_LINE_STATUS) {
2110                         if (copy_to_user(argp, &irq_event, sizeof irq_event))
2111                                 goto out;
2112                 }
2113
2114                 r = 0;
2115                 break;
2116         }
2117 #endif
2118         default:
2119                 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2120                 if (r == -ENOTTY)
2121                         r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2122         }
2123 out:
2124         return r;
2125 }
2126
2127 #ifdef CONFIG_COMPAT
2128 struct compat_kvm_dirty_log {
2129         __u32 slot;
2130         __u32 padding1;
2131         union {
2132                 compat_uptr_t dirty_bitmap; /* one bit per page */
2133                 __u64 padding2;
2134         };
2135 };
2136
2137 static long kvm_vm_compat_ioctl(struct file *filp,
2138                            unsigned int ioctl, unsigned long arg)
2139 {
2140         struct kvm *kvm = filp->private_data;
2141         int r;
2142
2143         if (kvm->mm != current->mm)
2144                 return -EIO;
2145         switch (ioctl) {
2146         case KVM_GET_DIRTY_LOG: {
2147                 struct compat_kvm_dirty_log compat_log;
2148                 struct kvm_dirty_log log;
2149
2150                 r = -EFAULT;
2151                 if (copy_from_user(&compat_log, (void __user *)arg,
2152                                    sizeof(compat_log)))
2153                         goto out;
2154                 log.slot         = compat_log.slot;
2155                 log.padding1     = compat_log.padding1;
2156                 log.padding2     = compat_log.padding2;
2157                 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2158
2159                 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2160                 if (r)
2161                         goto out;
2162                 break;
2163         }
2164         default:
2165                 r = kvm_vm_ioctl(filp, ioctl, arg);
2166         }
2167
2168 out:
2169         return r;
2170 }
2171 #endif
2172
2173 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2174 {
2175         struct page *page[1];
2176         unsigned long addr;
2177         int npages;
2178         gfn_t gfn = vmf->pgoff;
2179         struct kvm *kvm = vma->vm_file->private_data;
2180
2181         addr = gfn_to_hva(kvm, gfn);
2182         if (kvm_is_error_hva(addr))
2183                 return VM_FAULT_SIGBUS;
2184
2185         npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2186                                 NULL);
2187         if (unlikely(npages != 1))
2188                 return VM_FAULT_SIGBUS;
2189
2190         vmf->page = page[0];
2191         return 0;
2192 }
2193
2194 static const struct vm_operations_struct kvm_vm_vm_ops = {
2195         .fault = kvm_vm_fault,
2196 };
2197
2198 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2199 {
2200         vma->vm_ops = &kvm_vm_vm_ops;
2201         return 0;
2202 }
2203
2204 static struct file_operations kvm_vm_fops = {
2205         .release        = kvm_vm_release,
2206         .unlocked_ioctl = kvm_vm_ioctl,
2207 #ifdef CONFIG_COMPAT
2208         .compat_ioctl   = kvm_vm_compat_ioctl,
2209 #endif
2210         .mmap           = kvm_vm_mmap,
2211         .llseek         = noop_llseek,
2212 };
2213
2214 static int kvm_dev_ioctl_create_vm(unsigned long type)
2215 {
2216         int r;
2217         struct kvm *kvm;
2218
2219         kvm = kvm_create_vm(type);
2220         if (IS_ERR(kvm))
2221                 return PTR_ERR(kvm);
2222 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2223         r = kvm_coalesced_mmio_init(kvm);
2224         if (r < 0) {
2225                 kvm_put_kvm(kvm);
2226                 return r;
2227         }
2228 #endif
2229         r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2230         if (r < 0)
2231                 kvm_put_kvm(kvm);
2232
2233         return r;
2234 }
2235
2236 static long kvm_dev_ioctl_check_extension_generic(long arg)
2237 {
2238         switch (arg) {
2239         case KVM_CAP_USER_MEMORY:
2240         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2241         case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2242 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2243         case KVM_CAP_SET_BOOT_CPU_ID:
2244 #endif
2245         case KVM_CAP_INTERNAL_ERROR_DATA:
2246 #ifdef CONFIG_HAVE_KVM_MSI
2247         case KVM_CAP_SIGNAL_MSI:
2248 #endif
2249                 return 1;
2250 #ifdef KVM_CAP_IRQ_ROUTING
2251         case KVM_CAP_IRQ_ROUTING:
2252                 return KVM_MAX_IRQ_ROUTES;
2253 #endif
2254         default:
2255                 break;
2256         }
2257         return kvm_dev_ioctl_check_extension(arg);
2258 }
2259
2260 static long kvm_dev_ioctl(struct file *filp,
2261                           unsigned int ioctl, unsigned long arg)
2262 {
2263         long r = -EINVAL;
2264
2265         switch (ioctl) {
2266         case KVM_GET_API_VERSION:
2267                 r = -EINVAL;
2268                 if (arg)
2269                         goto out;
2270                 r = KVM_API_VERSION;
2271                 break;
2272         case KVM_CREATE_VM:
2273                 r = kvm_dev_ioctl_create_vm(arg);
2274                 break;
2275         case KVM_CHECK_EXTENSION:
2276                 r = kvm_dev_ioctl_check_extension_generic(arg);
2277                 break;
2278         case KVM_GET_VCPU_MMAP_SIZE:
2279                 r = -EINVAL;
2280                 if (arg)
2281                         goto out;
2282                 r = PAGE_SIZE;     /* struct kvm_run */
2283 #ifdef CONFIG_X86
2284                 r += PAGE_SIZE;    /* pio data page */
2285 #endif
2286 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2287                 r += PAGE_SIZE;    /* coalesced mmio ring page */
2288 #endif
2289                 break;
2290         case KVM_TRACE_ENABLE:
2291         case KVM_TRACE_PAUSE:
2292         case KVM_TRACE_DISABLE:
2293                 r = -EOPNOTSUPP;
2294                 break;
2295         default:
2296                 return kvm_arch_dev_ioctl(filp, ioctl, arg);
2297         }
2298 out:
2299         return r;
2300 }
2301
2302 static struct file_operations kvm_chardev_ops = {
2303         .unlocked_ioctl = kvm_dev_ioctl,
2304         .compat_ioctl   = kvm_dev_ioctl,
2305         .llseek         = noop_llseek,
2306 };
2307
2308 static struct miscdevice kvm_dev = {
2309         KVM_MINOR,
2310         "kvm",
2311         &kvm_chardev_ops,
2312 };
2313
2314 static void hardware_enable_nolock(void *junk)
2315 {
2316         int cpu = raw_smp_processor_id();
2317         int r;
2318
2319         if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2320                 return;
2321
2322         cpumask_set_cpu(cpu, cpus_hardware_enabled);
2323
2324         r = kvm_arch_hardware_enable(NULL);
2325
2326         if (r) {
2327                 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2328                 atomic_inc(&hardware_enable_failed);
2329                 printk(KERN_INFO "kvm: enabling virtualization on "
2330                                  "CPU%d failed\n", cpu);
2331         }
2332 }
2333
2334 static void hardware_enable(void *junk)
2335 {
2336         raw_spin_lock(&kvm_lock);
2337         hardware_enable_nolock(junk);
2338         raw_spin_unlock(&kvm_lock);
2339 }
2340
2341 static void hardware_disable_nolock(void *junk)
2342 {
2343         int cpu = raw_smp_processor_id();
2344
2345         if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2346                 return;
2347         cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2348         kvm_arch_hardware_disable(NULL);
2349 }
2350
2351 static void hardware_disable(void *junk)
2352 {
2353         raw_spin_lock(&kvm_lock);
2354         hardware_disable_nolock(junk);
2355         raw_spin_unlock(&kvm_lock);
2356 }
2357
2358 static void hardware_disable_all_nolock(void)
2359 {
2360         BUG_ON(!kvm_usage_count);
2361
2362         kvm_usage_count--;
2363         if (!kvm_usage_count)
2364                 on_each_cpu(hardware_disable_nolock, NULL, 1);
2365 }
2366
2367 static void hardware_disable_all(void)
2368 {
2369         raw_spin_lock(&kvm_lock);
2370         hardware_disable_all_nolock();
2371         raw_spin_unlock(&kvm_lock);
2372 }
2373
2374 static int hardware_enable_all(void)
2375 {
2376         int r = 0;
2377
2378         raw_spin_lock(&kvm_lock);
2379
2380         kvm_usage_count++;
2381         if (kvm_usage_count == 1) {
2382                 atomic_set(&hardware_enable_failed, 0);
2383                 on_each_cpu(hardware_enable_nolock, NULL, 1);
2384
2385                 if (atomic_read(&hardware_enable_failed)) {
2386                         hardware_disable_all_nolock();
2387                         r = -EBUSY;
2388                 }
2389         }
2390
2391         raw_spin_unlock(&kvm_lock);
2392
2393         return r;
2394 }
2395
2396 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2397                            void *v)
2398 {
2399         int cpu = (long)v;
2400
2401         if (!kvm_usage_count)
2402                 return NOTIFY_OK;
2403
2404         val &= ~CPU_TASKS_FROZEN;
2405         switch (val) {
2406         case CPU_DYING:
2407                 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2408                        cpu);
2409                 hardware_disable(NULL);
2410                 break;
2411         case CPU_STARTING:
2412                 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2413                        cpu);
2414                 hardware_enable(NULL);
2415                 break;
2416         }
2417         return NOTIFY_OK;
2418 }
2419
2420
2421 asmlinkage void kvm_spurious_fault(void)
2422 {
2423         /* Fault while not rebooting.  We want the trace. */
2424         BUG();
2425 }
2426 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2427
2428 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2429                       void *v)
2430 {
2431         /*
2432          * Some (well, at least mine) BIOSes hang on reboot if
2433          * in vmx root mode.
2434          *
2435          * And Intel TXT required VMX off for all cpu when system shutdown.
2436          */
2437         printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2438         kvm_rebooting = true;
2439         on_each_cpu(hardware_disable_nolock, NULL, 1);
2440         return NOTIFY_OK;
2441 }
2442
2443 static struct notifier_block kvm_reboot_notifier = {
2444         .notifier_call = kvm_reboot,
2445         .priority = 0,
2446 };
2447
2448 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2449 {
2450         int i;
2451
2452         for (i = 0; i < bus->dev_count; i++) {
2453                 struct kvm_io_device *pos = bus->range[i].dev;
2454
2455                 kvm_iodevice_destructor(pos);
2456         }
2457         kfree(bus);
2458 }
2459
2460 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2461 {
2462         const struct kvm_io_range *r1 = p1;
2463         const struct kvm_io_range *r2 = p2;
2464
2465         if (r1->addr < r2->addr)
2466                 return -1;
2467         if (r1->addr + r1->len > r2->addr + r2->len)
2468                 return 1;
2469         return 0;
2470 }
2471
2472 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2473                           gpa_t addr, int len)
2474 {
2475         bus->range[bus->dev_count++] = (struct kvm_io_range) {
2476                 .addr = addr,
2477                 .len = len,
2478                 .dev = dev,
2479         };
2480
2481         sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2482                 kvm_io_bus_sort_cmp, NULL);
2483
2484         return 0;
2485 }
2486
2487 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2488                              gpa_t addr, int len)
2489 {
2490         struct kvm_io_range *range, key;
2491         int off;
2492
2493         key = (struct kvm_io_range) {
2494                 .addr = addr,
2495                 .len = len,
2496         };
2497
2498         range = bsearch(&key, bus->range, bus->dev_count,
2499                         sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2500         if (range == NULL)
2501                 return -ENOENT;
2502
2503         off = range - bus->range;
2504
2505         while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2506                 off--;
2507
2508         return off;
2509 }
2510
2511 /* kvm_io_bus_write - called under kvm->slots_lock */
2512 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2513                      int len, const void *val)
2514 {
2515         int idx;
2516         struct kvm_io_bus *bus;
2517         struct kvm_io_range range;
2518
2519         range = (struct kvm_io_range) {
2520                 .addr = addr,
2521                 .len = len,
2522         };
2523
2524         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2525         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2526         if (idx < 0)
2527                 return -EOPNOTSUPP;
2528
2529         while (idx < bus->dev_count &&
2530                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2531                 if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2532                         return 0;
2533                 idx++;
2534         }
2535
2536         return -EOPNOTSUPP;
2537 }
2538
2539 /* kvm_io_bus_read - called under kvm->slots_lock */
2540 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2541                     int len, void *val)
2542 {
2543         int idx;
2544         struct kvm_io_bus *bus;
2545         struct kvm_io_range range;
2546
2547         range = (struct kvm_io_range) {
2548                 .addr = addr,
2549                 .len = len,
2550         };
2551
2552         bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2553         idx = kvm_io_bus_get_first_dev(bus, addr, len);
2554         if (idx < 0)
2555                 return -EOPNOTSUPP;
2556
2557         while (idx < bus->dev_count &&
2558                 kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2559                 if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2560                         return 0;
2561                 idx++;
2562         }
2563
2564         return -EOPNOTSUPP;
2565 }
2566
2567 /* Caller must hold slots_lock. */
2568 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2569                             int len, struct kvm_io_device *dev)
2570 {
2571         struct kvm_io_bus *new_bus, *bus;
2572
2573         bus = kvm->buses[bus_idx];
2574         if (bus->dev_count > NR_IOBUS_DEVS - 1)
2575                 return -ENOSPC;
2576
2577         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
2578                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2579         if (!new_bus)
2580                 return -ENOMEM;
2581         memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
2582                sizeof(struct kvm_io_range)));
2583         kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2584         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2585         synchronize_srcu_expedited(&kvm->srcu);
2586         kfree(bus);
2587
2588         return 0;
2589 }
2590
2591 /* Caller must hold slots_lock. */
2592 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2593                               struct kvm_io_device *dev)
2594 {
2595         int i, r;
2596         struct kvm_io_bus *new_bus, *bus;
2597
2598         bus = kvm->buses[bus_idx];
2599         r = -ENOENT;
2600         for (i = 0; i < bus->dev_count; i++)
2601                 if (bus->range[i].dev == dev) {
2602                         r = 0;
2603                         break;
2604                 }
2605
2606         if (r)
2607                 return r;
2608
2609         new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
2610                           sizeof(struct kvm_io_range)), GFP_KERNEL);
2611         if (!new_bus)
2612                 return -ENOMEM;
2613
2614         memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
2615         new_bus->dev_count--;
2616         memcpy(new_bus->range + i, bus->range + i + 1,
2617                (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
2618
2619         rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2620         synchronize_srcu_expedited(&kvm->srcu);
2621         kfree(bus);
2622         return r;
2623 }
2624
2625 static struct notifier_block kvm_cpu_notifier = {
2626         .notifier_call = kvm_cpu_hotplug,
2627 };
2628
2629 static int vm_stat_get(void *_offset, u64 *val)
2630 {
2631         unsigned offset = (long)_offset;
2632         struct kvm *kvm;
2633
2634         *val = 0;
2635         raw_spin_lock(&kvm_lock);
2636         list_for_each_entry(kvm, &vm_list, vm_list)
2637                 *val += *(u32 *)((void *)kvm + offset);
2638         raw_spin_unlock(&kvm_lock);
2639         return 0;
2640 }
2641
2642 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2643
2644 static int vcpu_stat_get(void *_offset, u64 *val)
2645 {
2646         unsigned offset = (long)_offset;
2647         struct kvm *kvm;
2648         struct kvm_vcpu *vcpu;
2649         int i;
2650
2651         *val = 0;
2652         raw_spin_lock(&kvm_lock);
2653         list_for_each_entry(kvm, &vm_list, vm_list)
2654                 kvm_for_each_vcpu(i, vcpu, kvm)
2655                         *val += *(u32 *)((void *)vcpu + offset);
2656
2657         raw_spin_unlock(&kvm_lock);
2658         return 0;
2659 }
2660
2661 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2662
2663 static const struct file_operations *stat_fops[] = {
2664         [KVM_STAT_VCPU] = &vcpu_stat_fops,
2665         [KVM_STAT_VM]   = &vm_stat_fops,
2666 };
2667
2668 static int kvm_init_debug(void)
2669 {
2670         int r = -EFAULT;
2671         struct kvm_stats_debugfs_item *p;
2672
2673         kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2674         if (kvm_debugfs_dir == NULL)
2675                 goto out;
2676
2677         for (p = debugfs_entries; p->name; ++p) {
2678                 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2679                                                 (void *)(long)p->offset,
2680                                                 stat_fops[p->kind]);
2681                 if (p->dentry == NULL)
2682                         goto out_dir;
2683         }
2684
2685         return 0;
2686
2687 out_dir:
2688         debugfs_remove_recursive(kvm_debugfs_dir);
2689 out:
2690         return r;
2691 }
2692
2693 static void kvm_exit_debug(void)
2694 {
2695         struct kvm_stats_debugfs_item *p;
2696
2697         for (p = debugfs_entries; p->name; ++p)
2698                 debugfs_remove(p->dentry);
2699         debugfs_remove(kvm_debugfs_dir);
2700 }
2701
2702 static int kvm_suspend(void)
2703 {
2704         if (kvm_usage_count)
2705                 hardware_disable_nolock(NULL);
2706         return 0;
2707 }
2708
2709 static void kvm_resume(void)
2710 {
2711         if (kvm_usage_count) {
2712                 WARN_ON(raw_spin_is_locked(&kvm_lock));
2713                 hardware_enable_nolock(NULL);
2714         }
2715 }
2716
2717 static struct syscore_ops kvm_syscore_ops = {
2718         .suspend = kvm_suspend,
2719         .resume = kvm_resume,
2720 };
2721
2722 static inline
2723 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2724 {
2725         return container_of(pn, struct kvm_vcpu, preempt_notifier);
2726 }
2727
2728 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2729 {
2730         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2731
2732         kvm_arch_vcpu_load(vcpu, cpu);
2733 }
2734
2735 static void kvm_sched_out(struct preempt_notifier *pn,
2736                           struct task_struct *next)
2737 {
2738         struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2739
2740         kvm_arch_vcpu_put(vcpu);
2741 }
2742
2743 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2744                   struct module *module)
2745 {
2746         int r;
2747         int cpu;
2748
2749         r = kvm_arch_init(opaque);
2750         if (r)
2751                 goto out_fail;
2752
2753         if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2754                 r = -ENOMEM;
2755                 goto out_free_0;
2756         }
2757
2758         r = kvm_arch_hardware_setup();
2759         if (r < 0)
2760                 goto out_free_0a;
2761
2762         for_each_online_cpu(cpu) {
2763                 smp_call_function_single(cpu,
2764                                 kvm_arch_check_processor_compat,
2765                                 &r, 1);
2766                 if (r < 0)
2767                         goto out_free_1;
2768         }
2769
2770         r = register_cpu_notifier(&kvm_cpu_notifier);
2771         if (r)
2772                 goto out_free_2;
2773         register_reboot_notifier(&kvm_reboot_notifier);
2774
2775         /* A kmem cache lets us meet the alignment requirements of fx_save. */
2776         if (!vcpu_align)
2777                 vcpu_align = __alignof__(struct kvm_vcpu);
2778         kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2779                                            0, NULL);
2780         if (!kvm_vcpu_cache) {
2781                 r = -ENOMEM;
2782                 goto out_free_3;
2783         }
2784
2785         r = kvm_async_pf_init();
2786         if (r)
2787                 goto out_free;
2788
2789         kvm_chardev_ops.owner = module;
2790         kvm_vm_fops.owner = module;
2791         kvm_vcpu_fops.owner = module;
2792
2793         r = misc_register(&kvm_dev);
2794         if (r) {
2795                 printk(KERN_ERR "kvm: misc device register failed\n");
2796                 goto out_unreg;
2797         }
2798
2799         register_syscore_ops(&kvm_syscore_ops);
2800
2801         kvm_preempt_ops.sched_in = kvm_sched_in;
2802         kvm_preempt_ops.sched_out = kvm_sched_out;
2803
2804         r = kvm_init_debug();
2805         if (r) {
2806                 printk(KERN_ERR "kvm: create debugfs files failed\n");
2807                 goto out_undebugfs;
2808         }
2809
2810         return 0;
2811
2812 out_undebugfs:
2813         unregister_syscore_ops(&kvm_syscore_ops);
2814 out_unreg:
2815         kvm_async_pf_deinit();
2816 out_free:
2817         kmem_cache_destroy(kvm_vcpu_cache);
2818 out_free_3:
2819         unregister_reboot_notifier(&kvm_reboot_notifier);
2820         unregister_cpu_notifier(&kvm_cpu_notifier);
2821 out_free_2:
2822 out_free_1:
2823         kvm_arch_hardware_unsetup();
2824 out_free_0a:
2825         free_cpumask_var(cpus_hardware_enabled);
2826 out_free_0:
2827         kvm_arch_exit();
2828 out_fail:
2829         return r;
2830 }
2831 EXPORT_SYMBOL_GPL(kvm_init);
2832
2833 void kvm_exit(void)
2834 {
2835         kvm_exit_debug();
2836         misc_deregister(&kvm_dev);
2837         kmem_cache_destroy(kvm_vcpu_cache);
2838         kvm_async_pf_deinit();
2839         unregister_syscore_ops(&kvm_syscore_ops);
2840         unregister_reboot_notifier(&kvm_reboot_notifier);
2841         unregister_cpu_notifier(&kvm_cpu_notifier);
2842         on_each_cpu(hardware_disable_nolock, NULL, 1);
2843         kvm_arch_hardware_unsetup();
2844         kvm_arch_exit();
2845         free_cpumask_var(cpus_hardware_enabled);
2846 }
2847 EXPORT_SYMBOL_GPL(kvm_exit);