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