linux/virt/kvm/kvm_main.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Kernel-based Virtual Machine driver for Linux
   4 *
   5 * This module enables machines with Intel VT-x extensions to run virtual
   6 * machines without emulation or binary translation.
   7 *
   8 * Copyright (C) 2006 Qumranet, Inc.
   9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  10 *
  11 * Authors:
  12 *   Avi Kivity   <avi@qumranet.com>
  13 *   Yaniv Kamay  <yaniv@qumranet.com>
  14 */
  15
  16#include <kvm/iodev.h>
  17
  18#include <linux/kvm_host.h>
  19#include <linux/kvm.h>
  20#include <linux/module.h>
  21#include <linux/errno.h>
  22#include <linux/percpu.h>
  23#include <linux/mm.h>
  24#include <linux/miscdevice.h>
  25#include <linux/vmalloc.h>
  26#include <linux/reboot.h>
  27#include <linux/debugfs.h>
  28#include <linux/highmem.h>
  29#include <linux/file.h>
  30#include <linux/syscore_ops.h>
  31#include <linux/cpu.h>
  32#include <linux/sched/signal.h>
  33#include <linux/sched/mm.h>
  34#include <linux/sched/stat.h>
  35#include <linux/cpumask.h>
  36#include <linux/smp.h>
  37#include <linux/anon_inodes.h>
  38#include <linux/profile.h>
  39#include <linux/kvm_para.h>
  40#include <linux/pagemap.h>
  41#include <linux/mman.h>
  42#include <linux/swap.h>
  43#include <linux/bitops.h>
  44#include <linux/spinlock.h>
  45#include <linux/compat.h>
  46#include <linux/srcu.h>
  47#include <linux/hugetlb.h>
  48#include <linux/slab.h>
  49#include <linux/sort.h>
  50#include <linux/bsearch.h>
  51#include <linux/io.h>
  52#include <linux/lockdep.h>
  53#include <linux/kthread.h>
  54
  55#include <asm/processor.h>
  56#include <asm/ioctl.h>
  57#include <linux/uaccess.h>
  58#include <asm/pgtable.h>
  59
  60#include "coalesced_mmio.h"
  61#include "async_pf.h"
  62#include "vfio.h"
  63
  64#define CREATE_TRACE_POINTS
  65#include <trace/events/kvm.h>
  66
  67/* Worst case buffer size needed for holding an integer. */
  68#define ITOA_MAX_LEN 12
  69
  70MODULE_AUTHOR("Qumranet");
  71MODULE_LICENSE("GPL");
  72
  73/* Architectures should define their poll value according to the halt latency */
  74unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
  75module_param(halt_poll_ns, uint, 0644);
  76EXPORT_SYMBOL_GPL(halt_poll_ns);
  77
  78/* Default doubles per-vcpu halt_poll_ns. */
  79unsigned int halt_poll_ns_grow = 2;
  80module_param(halt_poll_ns_grow, uint, 0644);
  81EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
  82
  83/* The start value to grow halt_poll_ns from */
  84unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
  85module_param(halt_poll_ns_grow_start, uint, 0644);
  86EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
  87
  88/* Default resets per-vcpu halt_poll_ns . */
  89unsigned int halt_poll_ns_shrink;
  90module_param(halt_poll_ns_shrink, uint, 0644);
  91EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
  92
  93/*
  94 * Ordering of locks:
  95 *
  96 *      kvm->lock --> kvm->slots_lock --> kvm->irq_lock
  97 */
  98
  99DEFINE_MUTEX(kvm_lock);
 100static DEFINE_RAW_SPINLOCK(kvm_count_lock);
 101LIST_HEAD(vm_list);
 102
 103static cpumask_var_t cpus_hardware_enabled;
 104static int kvm_usage_count;
 105static atomic_t hardware_enable_failed;
 106
 107struct kmem_cache *kvm_vcpu_cache;
 108EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
 109
 110static __read_mostly struct preempt_ops kvm_preempt_ops;
 111
 112struct dentry *kvm_debugfs_dir;
 113EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
 114
 115static int kvm_debugfs_num_entries;
 116static const struct file_operations *stat_fops_per_vm[];
 117
 118static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 119                           unsigned long arg);
 120#ifdef CONFIG_KVM_COMPAT
 121static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
 122                                  unsigned long arg);
 123#define KVM_COMPAT(c)   .compat_ioctl   = (c)
 124#else
 125/*
 126 * For architectures that don't implement a compat infrastructure,
 127 * adopt a double line of defense:
 128 * - Prevent a compat task from opening /dev/kvm
 129 * - If the open has been done by a 64bit task, and the KVM fd
 130 *   passed to a compat task, let the ioctls fail.
 131 */
 132static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
 133                                unsigned long arg) { return -EINVAL; }
 134
 135static int kvm_no_compat_open(struct inode *inode, struct file *file)
 136{
 137        return is_compat_task() ? -ENODEV : 0;
 138}
 139#define KVM_COMPAT(c)   .compat_ioctl   = kvm_no_compat_ioctl,  \
 140                        .open           = kvm_no_compat_open
 141#endif
 142static int hardware_enable_all(void);
 143static void hardware_disable_all(void);
 144
 145static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
 146
 147static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
 148
 149__visible bool kvm_rebooting;
 150EXPORT_SYMBOL_GPL(kvm_rebooting);
 151
 152static bool largepages_enabled = true;
 153
 154#define KVM_EVENT_CREATE_VM 0
 155#define KVM_EVENT_DESTROY_VM 1
 156static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
 157static unsigned long long kvm_createvm_count;
 158static unsigned long long kvm_active_vms;
 159
 160__weak int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
 161                unsigned long start, unsigned long end, bool blockable)
 162{
 163        return 0;
 164}
 165
 166bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
 167{
 168        /*
 169         * The metadata used by is_zone_device_page() to determine whether or
 170         * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
 171         * the device has been pinned, e.g. by get_user_pages().  WARN if the
 172         * page_count() is zero to help detect bad usage of this helper.
 173         */
 174        if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
 175                return false;
 176
 177        return is_zone_device_page(pfn_to_page(pfn));
 178}
 179
 180bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
 181{
 182        /*
 183         * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
 184         * perspective they are "normal" pages, albeit with slightly different
 185         * usage rules.
 186         */
 187        if (pfn_valid(pfn))
 188                return PageReserved(pfn_to_page(pfn)) &&
 189                       !kvm_is_zone_device_pfn(pfn);
 190
 191        return true;
 192}
 193
 194/*
 195 * Switches to specified vcpu, until a matching vcpu_put()
 196 */
 197void vcpu_load(struct kvm_vcpu *vcpu)
 198{
 199        int cpu = get_cpu();
 200        preempt_notifier_register(&vcpu->preempt_notifier);
 201        kvm_arch_vcpu_load(vcpu, cpu);
 202        put_cpu();
 203}
 204EXPORT_SYMBOL_GPL(vcpu_load);
 205
 206void vcpu_put(struct kvm_vcpu *vcpu)
 207{
 208        preempt_disable();
 209        kvm_arch_vcpu_put(vcpu);
 210        preempt_notifier_unregister(&vcpu->preempt_notifier);
 211        preempt_enable();
 212}
 213EXPORT_SYMBOL_GPL(vcpu_put);
 214
 215/* TODO: merge with kvm_arch_vcpu_should_kick */
 216static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
 217{
 218        int mode = kvm_vcpu_exiting_guest_mode(vcpu);
 219
 220        /*
 221         * We need to wait for the VCPU to reenable interrupts and get out of
 222         * READING_SHADOW_PAGE_TABLES mode.
 223         */
 224        if (req & KVM_REQUEST_WAIT)
 225                return mode != OUTSIDE_GUEST_MODE;
 226
 227        /*
 228         * Need to kick a running VCPU, but otherwise there is nothing to do.
 229         */
 230        return mode == IN_GUEST_MODE;
 231}
 232
 233static void ack_flush(void *_completed)
 234{
 235}
 236
 237static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
 238{
 239        if (unlikely(!cpus))
 240                cpus = cpu_online_mask;
 241
 242        if (cpumask_empty(cpus))
 243                return false;
 244
 245        smp_call_function_many(cpus, ack_flush, NULL, wait);
 246        return true;
 247}
 248
 249bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
 250                                 unsigned long *vcpu_bitmap, cpumask_var_t tmp)
 251{
 252        int i, cpu, me;
 253        struct kvm_vcpu *vcpu;
 254        bool called;
 255
 256        me = get_cpu();
 257
 258        kvm_for_each_vcpu(i, vcpu, kvm) {
 259                if (vcpu_bitmap && !test_bit(i, vcpu_bitmap))
 260                        continue;
 261
 262                kvm_make_request(req, vcpu);
 263                cpu = vcpu->cpu;
 264
 265                if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
 266                        continue;
 267
 268                if (tmp != NULL && cpu != -1 && cpu != me &&
 269                    kvm_request_needs_ipi(vcpu, req))
 270                        __cpumask_set_cpu(cpu, tmp);
 271        }
 272
 273        called = kvm_kick_many_cpus(tmp, !!(req & KVM_REQUEST_WAIT));
 274        put_cpu();
 275
 276        return called;
 277}
 278
 279bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
 280{
 281        cpumask_var_t cpus;
 282        bool called;
 283
 284        zalloc_cpumask_var(&cpus, GFP_ATOMIC);
 285
 286        called = kvm_make_vcpus_request_mask(kvm, req, NULL, cpus);
 287
 288        free_cpumask_var(cpus);
 289        return called;
 290}
 291
 292#ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
 293void kvm_flush_remote_tlbs(struct kvm *kvm)
 294{
 295        /*
 296         * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
 297         * kvm_make_all_cpus_request.
 298         */
 299        long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
 300
 301        /*
 302         * We want to publish modifications to the page tables before reading
 303         * mode. Pairs with a memory barrier in arch-specific code.
 304         * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
 305         * and smp_mb in walk_shadow_page_lockless_begin/end.
 306         * - powerpc: smp_mb in kvmppc_prepare_to_enter.
 307         *
 308         * There is already an smp_mb__after_atomic() before
 309         * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
 310         * barrier here.
 311         */
 312        if (!kvm_arch_flush_remote_tlb(kvm)
 313            || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 314                ++kvm->stat.remote_tlb_flush;
 315        cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
 316}
 317EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 318#endif
 319
 320void kvm_reload_remote_mmus(struct kvm *kvm)
 321{
 322        kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
 323}
 324
 325int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 326{
 327        struct page *page;
 328        int r;
 329
 330        mutex_init(&vcpu->mutex);
 331        vcpu->cpu = -1;
 332        vcpu->kvm = kvm;
 333        vcpu->vcpu_id = id;
 334        vcpu->pid = NULL;
 335        init_swait_queue_head(&vcpu->wq);
 336        kvm_async_pf_vcpu_init(vcpu);
 337
 338        vcpu->pre_pcpu = -1;
 339        INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
 340
 341        page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 342        if (!page) {
 343                r = -ENOMEM;
 344                goto fail;
 345        }
 346        vcpu->run = page_address(page);
 347
 348        kvm_vcpu_set_in_spin_loop(vcpu, false);
 349        kvm_vcpu_set_dy_eligible(vcpu, false);
 350        vcpu->preempted = false;
 351        vcpu->ready = false;
 352
 353        r = kvm_arch_vcpu_init(vcpu);
 354        if (r < 0)
 355                goto fail_free_run;
 356        return 0;
 357
 358fail_free_run:
 359        free_page((unsigned long)vcpu->run);
 360fail:
 361        return r;
 362}
 363EXPORT_SYMBOL_GPL(kvm_vcpu_init);
 364
 365void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
 366{
 367        /*
 368         * no need for rcu_read_lock as VCPU_RUN is the only place that
 369         * will change the vcpu->pid pointer and on uninit all file
 370         * descriptors are already gone.
 371         */
 372        put_pid(rcu_dereference_protected(vcpu->pid, 1));
 373        kvm_arch_vcpu_uninit(vcpu);
 374        free_page((unsigned long)vcpu->run);
 375}
 376EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
 377
 378#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 379static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 380{
 381        return container_of(mn, struct kvm, mmu_notifier);
 382}
 383
 384static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 385                                        struct mm_struct *mm,
 386                                        unsigned long address,
 387                                        pte_t pte)
 388{
 389        struct kvm *kvm = mmu_notifier_to_kvm(mn);
 390        int idx;
 391
 392        idx = srcu_read_lock(&kvm->srcu);
 393        spin_lock(&kvm->mmu_lock);
 394        kvm->mmu_notifier_seq++;
 395
 396        if (kvm_set_spte_hva(kvm, address, pte))
 397                kvm_flush_remote_tlbs(kvm);
 398
 399        spin_unlock(&kvm->mmu_lock);
 400        srcu_read_unlock(&kvm->srcu, idx);
 401}
 402
 403static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 404                                        const struct mmu_notifier_range *range)
 405{
 406        struct kvm *kvm = mmu_notifier_to_kvm(mn);
 407        int need_tlb_flush = 0, idx;
 408        int ret;
 409
 410        idx = srcu_read_lock(&kvm->srcu);
 411        spin_lock(&kvm->mmu_lock);
 412        /*
 413         * The count increase must become visible at unlock time as no
 414         * spte can be established without taking the mmu_lock and
 415         * count is also read inside the mmu_lock critical section.
 416         */
 417        kvm->mmu_notifier_count++;
 418        need_tlb_flush = kvm_unmap_hva_range(kvm, range->start, range->end);
 419        need_tlb_flush |= kvm->tlbs_dirty;
 420        /* we've to flush the tlb before the pages can be freed */
 421        if (need_tlb_flush)
 422                kvm_flush_remote_tlbs(kvm);
 423
 424        spin_unlock(&kvm->mmu_lock);
 425
 426        ret = kvm_arch_mmu_notifier_invalidate_range(kvm, range->start,
 427                                        range->end,
 428                                        mmu_notifier_range_blockable(range));
 429
 430        srcu_read_unlock(&kvm->srcu, idx);
 431
 432        return ret;
 433}
 434
 435static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 436                                        const struct mmu_notifier_range *range)
 437{
 438        struct kvm *kvm = mmu_notifier_to_kvm(mn);
 439
 440        spin_lock(&kvm->mmu_lock);
 441        /*
 442         * This sequence increase will notify the kvm page fault that
 443         * the page that is going to be mapped in the spte could have
 444         * been freed.
 445         */
 446        kvm->mmu_notifier_seq++;
 447        smp_wmb();
 448        /*
 449         * The above sequence increase must be visible before the
 450         * below count decrease, which is ensured by the smp_wmb above
 451         * in conjunction with the smp_rmb in mmu_notifier_retry().
 452         */
 453        kvm->mmu_notifier_count--;
 454        spin_unlock(&kvm->mmu_lock);
 455
 456        BUG_ON(kvm->mmu_notifier_count < 0);
 457}
 458
 459static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 460                                              struct mm_struct *mm,
 461                                              unsigned long start,
 462                                              unsigned long end)
 463{
 464        struct kvm *kvm = mmu_notifier_to_kvm(mn);
 465        int young, idx;
 466
 467        idx = srcu_read_lock(&kvm->srcu);
 468        spin_lock(&kvm->mmu_lock);
 469
 470        young = kvm_age_hva(kvm, start, end);
 471        if (young)
 472                kvm_flush_remote_tlbs(kvm);
 473
 474        spin_unlock(&kvm->mmu_lock);
 475        srcu_read_unlock(&kvm->srcu, idx);
 476
 477        return young;
 478}
 479
 480static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
 481                                        struct mm_struct *mm,
 482                                        unsigned long start,
 483                                        unsigned long end)
 484{
 485        struct kvm *kvm = mmu_notifier_to_kvm(mn);
 486        int young, idx;
 487
 488        idx = srcu_read_lock(&kvm->srcu);
 489        spin_lock(&kvm->mmu_lock);
 490        /*
 491         * Even though we do not flush TLB, this will still adversely
 492         * affect performance on pre-Haswell Intel EPT, where there is
 493         * no EPT Access Bit to clear so that we have to tear down EPT
 494         * tables instead. If we find this unacceptable, we can always
 495         * add a parameter to kvm_age_hva so that it effectively doesn't
 496         * do anything on clear_young.
 497         *
 498         * Also note that currently we never issue secondary TLB flushes
 499         * from clear_young, leaving this job up to the regular system
 500         * cadence. If we find this inaccurate, we might come up with a
 501         * more sophisticated heuristic later.
 502         */
 503        young = kvm_age_hva(kvm, start, end);
 504        spin_unlock(&kvm->mmu_lock);
 505        srcu_read_unlock(&kvm->srcu, idx);
 506
 507        return young;
 508}
 509
 510static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 511                                       struct mm_struct *mm,
 512                                       unsigned long address)
 513{
 514        struct kvm *kvm = mmu_notifier_to_kvm(mn);
 515        int young, idx;
 516
 517        idx = srcu_read_lock(&kvm->srcu);
 518        spin_lock(&kvm->mmu_lock);
 519        young = kvm_test_age_hva(kvm, address);
 520        spin_unlock(&kvm->mmu_lock);
 521        srcu_read_unlock(&kvm->srcu, idx);
 522
 523        return young;
 524}
 525
 526static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 527                                     struct mm_struct *mm)
 528{
 529        struct kvm *kvm = mmu_notifier_to_kvm(mn);
 530        int idx;
 531
 532        idx = srcu_read_lock(&kvm->srcu);
 533        kvm_arch_flush_shadow_all(kvm);
 534        srcu_read_unlock(&kvm->srcu, idx);
 535}
 536
 537static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 538        .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
 539        .invalidate_range_end   = kvm_mmu_notifier_invalidate_range_end,
 540        .clear_flush_young      = kvm_mmu_notifier_clear_flush_young,
 541        .clear_young            = kvm_mmu_notifier_clear_young,
 542        .test_young             = kvm_mmu_notifier_test_young,
 543        .change_pte             = kvm_mmu_notifier_change_pte,
 544        .release                = kvm_mmu_notifier_release,
 545};
 546
 547static int kvm_init_mmu_notifier(struct kvm *kvm)
 548{
 549        kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 550        return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 551}
 552
 553#else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 554
 555static int kvm_init_mmu_notifier(struct kvm *kvm)
 556{
 557        return 0;
 558}
 559
 560#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 561
 562static struct kvm_memslots *kvm_alloc_memslots(void)
 563{
 564        int i;
 565        struct kvm_memslots *slots;
 566
 567        slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
 568        if (!slots)
 569                return NULL;
 570
 571        for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
 572                slots->id_to_index[i] = slots->memslots[i].id = i;
 573
 574        return slots;
 575}
 576
 577static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 578{
 579        if (!memslot->dirty_bitmap)
 580                return;
 581
 582        kvfree(memslot->dirty_bitmap);
 583        memslot->dirty_bitmap = NULL;
 584}
 585
 586/*
 587 * Free any memory in @free but not in @dont.
 588 */
 589static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 590                              struct kvm_memory_slot *dont)
 591{
 592        if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
 593                kvm_destroy_dirty_bitmap(free);
 594
 595        kvm_arch_free_memslot(kvm, free, dont);
 596
 597        free->npages = 0;
 598}
 599
 600static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
 601{
 602        struct kvm_memory_slot *memslot;
 603
 604        if (!slots)
 605                return;
 606
 607        kvm_for_each_memslot(memslot, slots)
 608                kvm_free_memslot(kvm, memslot, NULL);
 609
 610        kvfree(slots);
 611}
 612
 613static void kvm_destroy_vm_debugfs(struct kvm *kvm)
 614{
 615        int i;
 616
 617        if (!kvm->debugfs_dentry)
 618                return;
 619
 620        debugfs_remove_recursive(kvm->debugfs_dentry);
 621
 622        if (kvm->debugfs_stat_data) {
 623                for (i = 0; i < kvm_debugfs_num_entries; i++)
 624                        kfree(kvm->debugfs_stat_data[i]);
 625                kfree(kvm->debugfs_stat_data);
 626        }
 627}
 628
 629static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
 630{
 631        char dir_name[ITOA_MAX_LEN * 2];
 632        struct kvm_stat_data *stat_data;
 633        struct kvm_stats_debugfs_item *p;
 634
 635        if (!debugfs_initialized())
 636                return 0;
 637
 638        snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
 639        kvm->debugfs_dentry = debugfs_create_dir(dir_name, kvm_debugfs_dir);
 640
 641        kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
 642                                         sizeof(*kvm->debugfs_stat_data),
 643                                         GFP_KERNEL_ACCOUNT);
 644        if (!kvm->debugfs_stat_data)
 645                return -ENOMEM;
 646
 647        for (p = debugfs_entries; p->name; p++) {
 648                stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
 649                if (!stat_data)
 650                        return -ENOMEM;
 651
 652                stat_data->kvm = kvm;
 653                stat_data->offset = p->offset;
 654                stat_data->mode = p->mode ? p->mode : 0644;
 655                kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
 656                debugfs_create_file(p->name, stat_data->mode, kvm->debugfs_dentry,
 657                                    stat_data, stat_fops_per_vm[p->kind]);
 658        }
 659        return 0;
 660}
 661
 662/*
 663 * Called after the VM is otherwise initialized, but just before adding it to
 664 * the vm_list.
 665 */
 666int __weak kvm_arch_post_init_vm(struct kvm *kvm)
 667{
 668        return 0;
 669}
 670
 671/*
 672 * Called just after removing the VM from the vm_list, but before doing any
 673 * other destruction.
 674 */
 675void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
 676{
 677}
 678
 679static struct kvm *kvm_create_vm(unsigned long type)
 680{
 681        struct kvm *kvm = kvm_arch_alloc_vm();
 682        int r = -ENOMEM;
 683        int i;
 684
 685        if (!kvm)
 686                return ERR_PTR(-ENOMEM);
 687
 688        spin_lock_init(&kvm->mmu_lock);
 689        mmgrab(current->mm);
 690        kvm->mm = current->mm;
 691        kvm_eventfd_init(kvm);
 692        mutex_init(&kvm->lock);
 693        mutex_init(&kvm->irq_lock);
 694        mutex_init(&kvm->slots_lock);
 695        INIT_LIST_HEAD(&kvm->devices);
 696
 697        BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
 698
 699        if (init_srcu_struct(&kvm->srcu))
 700                goto out_err_no_srcu;
 701        if (init_srcu_struct(&kvm->irq_srcu))
 702                goto out_err_no_irq_srcu;
 703
 704        refcount_set(&kvm->users_count, 1);
 705        for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
 706                struct kvm_memslots *slots = kvm_alloc_memslots();
 707
 708                if (!slots)
 709                        goto out_err_no_arch_destroy_vm;
 710                /* Generations must be different for each address space. */
 711                slots->generation = i;
 712                rcu_assign_pointer(kvm->memslots[i], slots);
 713        }
 714
 715        for (i = 0; i < KVM_NR_BUSES; i++) {
 716                rcu_assign_pointer(kvm->buses[i],
 717                        kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
 718                if (!kvm->buses[i])
 719                        goto out_err_no_arch_destroy_vm;
 720        }
 721
 722        r = kvm_arch_init_vm(kvm, type);
 723        if (r)
 724                goto out_err_no_arch_destroy_vm;
 725
 726        r = hardware_enable_all();
 727        if (r)
 728                goto out_err_no_disable;
 729
 730#ifdef CONFIG_HAVE_KVM_IRQFD
 731        INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
 732#endif
 733
 734        r = kvm_init_mmu_notifier(kvm);
 735        if (r)
 736                goto out_err_no_mmu_notifier;
 737
 738        r = kvm_arch_post_init_vm(kvm);
 739        if (r)
 740                goto out_err;
 741
 742        mutex_lock(&kvm_lock);
 743        list_add(&kvm->vm_list, &vm_list);
 744        mutex_unlock(&kvm_lock);
 745
 746        preempt_notifier_inc();
 747
 748        return kvm;
 749
 750out_err:
 751#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 752        if (kvm->mmu_notifier.ops)
 753                mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
 754#endif
 755out_err_no_mmu_notifier:
 756        hardware_disable_all();
 757out_err_no_disable:
 758        kvm_arch_destroy_vm(kvm);
 759out_err_no_arch_destroy_vm:
 760        WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
 761        for (i = 0; i < KVM_NR_BUSES; i++)
 762                kfree(kvm_get_bus(kvm, i));
 763        for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
 764                kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
 765        cleanup_srcu_struct(&kvm->irq_srcu);
 766out_err_no_irq_srcu:
 767        cleanup_srcu_struct(&kvm->srcu);
 768out_err_no_srcu:
 769        kvm_arch_free_vm(kvm);
 770        mmdrop(current->mm);
 771        return ERR_PTR(r);
 772}
 773
 774static void kvm_destroy_devices(struct kvm *kvm)
 775{
 776        struct kvm_device *dev, *tmp;
 777
 778        /*
 779         * We do not need to take the kvm->lock here, because nobody else
 780         * has a reference to the struct kvm at this point and therefore
 781         * cannot access the devices list anyhow.
 782         */
 783        list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
 784                list_del(&dev->vm_node);
 785                dev->ops->destroy(dev);
 786        }
 787}
 788
 789static void kvm_destroy_vm(struct kvm *kvm)
 790{
 791        int i;
 792        struct mm_struct *mm = kvm->mm;
 793
 794        kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
 795        kvm_destroy_vm_debugfs(kvm);
 796        kvm_arch_sync_events(kvm);
 797        mutex_lock(&kvm_lock);
 798        list_del(&kvm->vm_list);
 799        mutex_unlock(&kvm_lock);
 800        kvm_arch_pre_destroy_vm(kvm);
 801
 802        kvm_free_irq_routing(kvm);
 803        for (i = 0; i < KVM_NR_BUSES; i++) {
 804                struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
 805
 806                if (bus)
 807                        kvm_io_bus_destroy(bus);
 808                kvm->buses[i] = NULL;
 809        }
 810        kvm_coalesced_mmio_free(kvm);
 811#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 812        mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
 813#else
 814        kvm_arch_flush_shadow_all(kvm);
 815#endif
 816        kvm_arch_destroy_vm(kvm);
 817        kvm_destroy_devices(kvm);
 818        for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
 819                kvm_free_memslots(kvm, __kvm_memslots(kvm, i));
 820        cleanup_srcu_struct(&kvm->irq_srcu);
 821        cleanup_srcu_struct(&kvm->srcu);
 822        kvm_arch_free_vm(kvm);
 823        preempt_notifier_dec();
 824        hardware_disable_all();
 825        mmdrop(mm);
 826}
 827
 828void kvm_get_kvm(struct kvm *kvm)
 829{
 830        refcount_inc(&kvm->users_count);
 831}
 832EXPORT_SYMBOL_GPL(kvm_get_kvm);
 833
 834void kvm_put_kvm(struct kvm *kvm)
 835{
 836        if (refcount_dec_and_test(&kvm->users_count))
 837                kvm_destroy_vm(kvm);
 838}
 839EXPORT_SYMBOL_GPL(kvm_put_kvm);
 840
 841
 842static int kvm_vm_release(struct inode *inode, struct file *filp)
 843{
 844        struct kvm *kvm = filp->private_data;
 845
 846        kvm_irqfd_release(kvm);
 847
 848        kvm_put_kvm(kvm);
 849        return 0;
 850}
 851
 852/*
 853 * Allocation size is twice as large as the actual dirty bitmap size.
 854 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
 855 */
 856static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
 857{
 858        unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
 859
 860        memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
 861        if (!memslot->dirty_bitmap)
 862                return -ENOMEM;
 863
 864        return 0;
 865}
 866
 867/*
 868 * Insert memslot and re-sort memslots based on their GFN,
 869 * so binary search could be used to lookup GFN.
 870 * Sorting algorithm takes advantage of having initially
 871 * sorted array and known changed memslot position.
 872 */
 873static void update_memslots(struct kvm_memslots *slots,
 874                            struct kvm_memory_slot *new,
 875                            enum kvm_mr_change change)
 876{
 877        int id = new->id;
 878        int i = slots->id_to_index[id];
 879        struct kvm_memory_slot *mslots = slots->memslots;
 880
 881        WARN_ON(mslots[i].id != id);
 882        switch (change) {
 883        case KVM_MR_CREATE:
 884                slots->used_slots++;
 885                WARN_ON(mslots[i].npages || !new->npages);
 886                break;
 887        case KVM_MR_DELETE:
 888                slots->used_slots--;
 889                WARN_ON(new->npages || !mslots[i].npages);
 890                break;
 891        default:
 892                break;
 893        }
 894
 895        while (i < KVM_MEM_SLOTS_NUM - 1 &&
 896               new->base_gfn <= mslots[i + 1].base_gfn) {
 897                if (!mslots[i + 1].npages)
 898                        break;
 899                mslots[i] = mslots[i + 1];
 900                slots->id_to_index[mslots[i].id] = i;
 901                i++;
 902        }
 903
 904        /*
 905         * The ">=" is needed when creating a slot with base_gfn == 0,
 906         * so that it moves before all those with base_gfn == npages == 0.
 907         *
 908         * On the other hand, if new->npages is zero, the above loop has
 909         * already left i pointing to the beginning of the empty part of
 910         * mslots, and the ">=" would move the hole backwards in this
 911         * case---which is wrong.  So skip the loop when deleting a slot.
 912         */
 913        if (new->npages) {
 914                while (i > 0 &&
 915                       new->base_gfn >= mslots[i - 1].base_gfn) {
 916                        mslots[i] = mslots[i - 1];
 917                        slots->id_to_index[mslots[i].id] = i;
 918                        i--;
 919                }
 920        } else
 921                WARN_ON_ONCE(i != slots->used_slots);
 922
 923        mslots[i] = *new;
 924        slots->id_to_index[mslots[i].id] = i;
 925}
 926
 927static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
 928{
 929        u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
 930
 931#ifdef __KVM_HAVE_READONLY_MEM
 932        valid_flags |= KVM_MEM_READONLY;
 933#endif
 934
 935        if (mem->flags & ~valid_flags)
 936                return -EINVAL;
 937
 938        return 0;
 939}
 940
 941static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
 942                int as_id, struct kvm_memslots *slots)
 943{
 944        struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
 945        u64 gen = old_memslots->generation;
 946
 947        WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
 948        slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
 949
 950        rcu_assign_pointer(kvm->memslots[as_id], slots);
 951        synchronize_srcu_expedited(&kvm->srcu);
 952
 953        /*
 954         * Increment the new memslot generation a second time, dropping the
 955         * update in-progress flag and incrementing then generation based on
 956         * the number of address spaces.  This provides a unique and easily
 957         * identifiable generation number while the memslots are in flux.
 958         */
 959        gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
 960
 961        /*
 962         * Generations must be unique even across address spaces.  We do not need
 963         * a global counter for that, instead the generation space is evenly split
 964         * across address spaces.  For example, with two address spaces, address
 965         * space 0 will use generations 0, 2, 4, ... while address space 1 will
 966         * use generations 1, 3, 5, ...
 967         */
 968        gen += KVM_ADDRESS_SPACE_NUM;
 969
 970        kvm_arch_memslots_updated(kvm, gen);
 971
 972        slots->generation = gen;
 973
 974        return old_memslots;
 975}
 976
 977/*
 978 * Allocate some memory and give it an address in the guest physical address
 979 * space.
 980 *
 981 * Discontiguous memory is allowed, mostly for framebuffers.
 982 *
 983 * Must be called holding kvm->slots_lock for write.
 984 */
 985int __kvm_set_memory_region(struct kvm *kvm,
 986                            const struct kvm_userspace_memory_region *mem)
 987{
 988        int r;
 989        gfn_t base_gfn;
 990        unsigned long npages;
 991        struct kvm_memory_slot *slot;
 992        struct kvm_memory_slot old, new;
 993        struct kvm_memslots *slots = NULL, *old_memslots;
 994        int as_id, id;
 995        enum kvm_mr_change change;
 996
 997        r = check_memory_region_flags(mem);
 998        if (r)
 999                goto out;
1000
1001        r = -EINVAL;
1002        as_id = mem->slot >> 16;
1003        id = (u16)mem->slot;
1004
1005        /* General sanity checks */
1006        if (mem->memory_size & (PAGE_SIZE - 1))
1007                goto out;
1008        if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1009                goto out;
1010        /* We can read the guest memory with __xxx_user() later on. */
1011        if ((id < KVM_USER_MEM_SLOTS) &&
1012            ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1013             !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1014                        mem->memory_size)))
1015                goto out;
1016        if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1017                goto out;
1018        if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1019                goto out;
1020
1021        slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
1022        base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
1023        npages = mem->memory_size >> PAGE_SHIFT;
1024
1025        if (npages > KVM_MEM_MAX_NR_PAGES)
1026                goto out;
1027
1028        new = old = *slot;
1029
1030        new.id = id;
1031        new.base_gfn = base_gfn;
1032        new.npages = npages;
1033        new.flags = mem->flags;
1034
1035        if (npages) {
1036                if (!old.npages)
1037                        change = KVM_MR_CREATE;
1038                else { /* Modify an existing slot. */
1039                        if ((mem->userspace_addr != old.userspace_addr) ||
1040                            (npages != old.npages) ||
1041                            ((new.flags ^ old.flags) & KVM_MEM_READONLY))
1042                                goto out;
1043
1044                        if (base_gfn != old.base_gfn)
1045                                change = KVM_MR_MOVE;
1046                        else if (new.flags != old.flags)
1047                                change = KVM_MR_FLAGS_ONLY;
1048                        else { /* Nothing to change. */
1049                                r = 0;
1050                                goto out;
1051                        }
1052                }
1053        } else {
1054                if (!old.npages)
1055                        goto out;
1056
1057                change = KVM_MR_DELETE;
1058                new.base_gfn = 0;
1059                new.flags = 0;
1060        }
1061
1062        if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1063                /* Check for overlaps */
1064                r = -EEXIST;
1065                kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1066                        if (slot->id == id)
1067                                continue;
1068                        if (!((base_gfn + npages <= slot->base_gfn) ||
1069                              (base_gfn >= slot->base_gfn + slot->npages)))
1070                                goto out;
1071                }
1072        }
1073
1074        /* Free page dirty bitmap if unneeded */
1075        if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1076                new.dirty_bitmap = NULL;
1077
1078        r = -ENOMEM;
1079        if (change == KVM_MR_CREATE) {
1080                new.userspace_addr = mem->userspace_addr;
1081
1082                if (kvm_arch_create_memslot(kvm, &new, npages))
1083                        goto out_free;
1084        }
1085
1086        /* Allocate page dirty bitmap if needed */
1087        if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1088                if (kvm_create_dirty_bitmap(&new) < 0)
1089                        goto out_free;
1090        }
1091
1092        slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL_ACCOUNT);
1093        if (!slots)
1094                goto out_free;
1095        memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1096
1097        if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1098                slot = id_to_memslot(slots, id);
1099                slot->flags |= KVM_MEMSLOT_INVALID;
1100
1101                old_memslots = install_new_memslots(kvm, as_id, slots);
1102
1103                /* From this point no new shadow pages pointing to a deleted,
1104                 * or moved, memslot will be created.
1105                 *
1106                 * validation of sp->gfn happens in:
1107                 *      - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1108                 *      - kvm_is_visible_gfn (mmu_check_roots)
1109                 */
1110                kvm_arch_flush_shadow_memslot(kvm, slot);
1111
1112                /*
1113                 * We can re-use the old_memslots from above, the only difference
1114                 * from the currently installed memslots is the invalid flag.  This
1115                 * will get overwritten by update_memslots anyway.
1116                 */
1117                slots = old_memslots;
1118        }
1119
1120        r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1121        if (r)
1122                goto out_slots;
1123
1124        /* actual memory is freed via old in kvm_free_memslot below */
1125        if (change == KVM_MR_DELETE) {
1126                new.dirty_bitmap = NULL;
1127                memset(&new.arch, 0, sizeof(new.arch));
1128        }
1129
1130        update_memslots(slots, &new, change);
1131        old_memslots = install_new_memslots(kvm, as_id, slots);
1132
1133        kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1134
1135        kvm_free_memslot(kvm, &old, &new);
1136        kvfree(old_memslots);
1137        return 0;
1138
1139out_slots:
1140        kvfree(slots);
1141out_free:
1142        kvm_free_memslot(kvm, &new, &old);
1143out:
1144        return r;
1145}
1146EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1147
1148int kvm_set_memory_region(struct kvm *kvm,
1149                          const struct kvm_userspace_memory_region *mem)
1150{
1151        int r;
1152
1153        mutex_lock(&kvm->slots_lock);
1154        r = __kvm_set_memory_region(kvm, mem);
1155        mutex_unlock(&kvm->slots_lock);
1156        return r;
1157}
1158EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1159
1160static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1161                                          struct kvm_userspace_memory_region *mem)
1162{
1163        if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1164                return -EINVAL;
1165
1166        return kvm_set_memory_region(kvm, mem);
1167}
1168
1169int kvm_get_dirty_log(struct kvm *kvm,
1170                        struct kvm_dirty_log *log, int *is_dirty)
1171{
1172        struct kvm_memslots *slots;
1173        struct kvm_memory_slot *memslot;
1174        int i, as_id, id;
1175        unsigned long n;
1176        unsigned long any = 0;
1177
1178        as_id = log->slot >> 16;
1179        id = (u16)log->slot;
1180        if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1181                return -EINVAL;
1182
1183        slots = __kvm_memslots(kvm, as_id);
1184        memslot = id_to_memslot(slots, id);
1185        if (!memslot->dirty_bitmap)
1186                return -ENOENT;
1187
1188        n = kvm_dirty_bitmap_bytes(memslot);
1189
1190        for (i = 0; !any && i < n/sizeof(long); ++i)
1191                any = memslot->dirty_bitmap[i];
1192
1193        if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1194                return -EFAULT;
1195
1196        if (any)
1197                *is_dirty = 1;
1198        return 0;
1199}
1200EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1201
1202#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1203/**
1204 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1205 *      and reenable dirty page tracking for the corresponding pages.
1206 * @kvm:        pointer to kvm instance
1207 * @log:        slot id and address to which we copy the log
1208 * @flush:      true if TLB flush is needed by caller
1209 *
1210 * We need to keep it in mind that VCPU threads can write to the bitmap
1211 * concurrently. So, to avoid losing track of dirty pages we keep the
1212 * following order:
1213 *
1214 *    1. Take a snapshot of the bit and clear it if needed.
1215 *    2. Write protect the corresponding page.
1216 *    3. Copy the snapshot to the userspace.
1217 *    4. Upon return caller flushes TLB's if needed.
1218 *
1219 * Between 2 and 4, the guest may write to the page using the remaining TLB
1220 * entry.  This is not a problem because the page is reported dirty using
1221 * the snapshot taken before and step 4 ensures that writes done after
1222 * exiting to userspace will be logged for the next call.
1223 *
1224 */
1225int kvm_get_dirty_log_protect(struct kvm *kvm,
1226                        struct kvm_dirty_log *log, bool *flush)
1227{
1228        struct kvm_memslots *slots;
1229        struct kvm_memory_slot *memslot;
1230        int i, as_id, id;
1231        unsigned long n;
1232        unsigned long *dirty_bitmap;
1233        unsigned long *dirty_bitmap_buffer;
1234
1235        as_id = log->slot >> 16;
1236        id = (u16)log->slot;
1237        if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1238                return -EINVAL;
1239
1240        slots = __kvm_memslots(kvm, as_id);
1241        memslot = id_to_memslot(slots, id);
1242
1243        dirty_bitmap = memslot->dirty_bitmap;
1244        if (!dirty_bitmap)
1245                return -ENOENT;
1246
1247        n = kvm_dirty_bitmap_bytes(memslot);
1248        *flush = false;
1249        if (kvm->manual_dirty_log_protect) {
1250                /*
1251                 * Unlike kvm_get_dirty_log, we always return false in *flush,
1252                 * because no flush is needed until KVM_CLEAR_DIRTY_LOG.  There
1253                 * is some code duplication between this function and
1254                 * kvm_get_dirty_log, but hopefully all architecture
1255                 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
1256                 * can be eliminated.
1257                 */
1258                dirty_bitmap_buffer = dirty_bitmap;
1259        } else {
1260                dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1261                memset(dirty_bitmap_buffer, 0, n);
1262
1263                spin_lock(&kvm->mmu_lock);
1264                for (i = 0; i < n / sizeof(long); i++) {
1265                        unsigned long mask;
1266                        gfn_t offset;
1267
1268                        if (!dirty_bitmap[i])
1269                                continue;
1270
1271                        *flush = true;
1272                        mask = xchg(&dirty_bitmap[i], 0);
1273                        dirty_bitmap_buffer[i] = mask;
1274
1275                        offset = i * BITS_PER_LONG;
1276                        kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1277                                                                offset, mask);
1278                }
1279                spin_unlock(&kvm->mmu_lock);
1280        }
1281
1282        if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1283                return -EFAULT;
1284        return 0;
1285}
1286EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1287
1288/**
1289 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
1290 *      and reenable dirty page tracking for the corresponding pages.
1291 * @kvm:        pointer to kvm instance
1292 * @log:        slot id and address from which to fetch the bitmap of dirty pages
1293 * @flush:      true if TLB flush is needed by caller
1294 */
1295int kvm_clear_dirty_log_protect(struct kvm *kvm,
1296                                struct kvm_clear_dirty_log *log, bool *flush)
1297{
1298        struct kvm_memslots *slots;
1299        struct kvm_memory_slot *memslot;
1300        int as_id, id;
1301        gfn_t offset;
1302        unsigned long i, n;
1303        unsigned long *dirty_bitmap;
1304        unsigned long *dirty_bitmap_buffer;
1305
1306        as_id = log->slot >> 16;
1307        id = (u16)log->slot;
1308        if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1309                return -EINVAL;
1310
1311        if (log->first_page & 63)
1312                return -EINVAL;
1313
1314        slots = __kvm_memslots(kvm, as_id);
1315        memslot = id_to_memslot(slots, id);
1316
1317        dirty_bitmap = memslot->dirty_bitmap;
1318        if (!dirty_bitmap)
1319                return -ENOENT;
1320
1321        n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
1322
1323        if (log->first_page > memslot->npages ||
1324            log->num_pages > memslot->npages - log->first_page ||
1325            (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
1326            return -EINVAL;
1327
1328        *flush = false;
1329        dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
1330        if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
1331                return -EFAULT;
1332
1333        spin_lock(&kvm->mmu_lock);
1334        for (offset = log->first_page, i = offset / BITS_PER_LONG,
1335                 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
1336             i++, offset += BITS_PER_LONG) {
1337                unsigned long mask = *dirty_bitmap_buffer++;
1338                atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
1339                if (!mask)
1340                        continue;
1341
1342                mask &= atomic_long_fetch_andnot(mask, p);
1343
1344                /*
1345                 * mask contains the bits that really have been cleared.  This
1346                 * never includes any bits beyond the length of the memslot (if
1347                 * the length is not aligned to 64 pages), therefore it is not
1348                 * a problem if userspace sets them in log->dirty_bitmap.
1349                */
1350                if (mask) {
1351                        *flush = true;
1352                        kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1353                                                                offset, mask);
1354                }
1355        }
1356        spin_unlock(&kvm->mmu_lock);
1357
1358        return 0;
1359}
1360EXPORT_SYMBOL_GPL(kvm_clear_dirty_log_protect);
1361#endif
1362
1363bool kvm_largepages_enabled(void)
1364{
1365        return largepages_enabled;
1366}
1367
1368void kvm_disable_largepages(void)
1369{
1370        largepages_enabled = false;
1371}
1372EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1373
1374struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1375{
1376        return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1377}
1378EXPORT_SYMBOL_GPL(gfn_to_memslot);
1379
1380struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1381{
1382        return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1383}
1384
1385bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1386{
1387        struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1388
1389        if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1390              memslot->flags & KVM_MEMSLOT_INVALID)
1391                return false;
1392
1393        return true;
1394}
1395EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1396
1397unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1398{
1399        struct vm_area_struct *vma;
1400        unsigned long addr, size;
1401
1402        size = PAGE_SIZE;
1403
1404        addr = gfn_to_hva(kvm, gfn);
1405        if (kvm_is_error_hva(addr))
1406                return PAGE_SIZE;
1407
1408        down_read(&current->mm->mmap_sem);
1409        vma = find_vma(current->mm, addr);
1410        if (!vma)
1411                goto out;
1412
1413        size = vma_kernel_pagesize(vma);
1414
1415out:
1416        up_read(&current->mm->mmap_sem);
1417
1418        return size;
1419}
1420
1421static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1422{
1423        return slot->flags & KVM_MEM_READONLY;
1424}
1425
1426static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1427                                       gfn_t *nr_pages, bool write)
1428{
1429        if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1430                return KVM_HVA_ERR_BAD;
1431
1432        if (memslot_is_readonly(slot) && write)
1433                return KVM_HVA_ERR_RO_BAD;
1434
1435        if (nr_pages)
1436                *nr_pages = slot->npages - (gfn - slot->base_gfn);
1437
1438        return __gfn_to_hva_memslot(slot, gfn);
1439}
1440
1441static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1442                                     gfn_t *nr_pages)
1443{
1444        return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1445}
1446
1447unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1448                                        gfn_t gfn)
1449{
1450        return gfn_to_hva_many(slot, gfn, NULL);
1451}
1452EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1453
1454unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1455{
1456        return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1457}
1458EXPORT_SYMBOL_GPL(gfn_to_hva);
1459
1460unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1461{
1462        return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1463}
1464EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1465
1466/*
1467 * Return the hva of a @gfn and the R/W attribute if possible.
1468 *
1469 * @slot: the kvm_memory_slot which contains @gfn
1470 * @gfn: the gfn to be translated
1471 * @writable: used to return the read/write attribute of the @slot if the hva
1472 * is valid and @writable is not NULL
1473 */
1474unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1475                                      gfn_t gfn, bool *writable)
1476{
1477        unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1478
1479        if (!kvm_is_error_hva(hva) && writable)
1480                *writable = !memslot_is_readonly(slot);
1481
1482        return hva;
1483}
1484
1485unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1486{
1487        struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1488
1489        return gfn_to_hva_memslot_prot(slot, gfn, writable);
1490}
1491
1492unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1493{
1494        struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1495
1496        return gfn_to_hva_memslot_prot(slot, gfn, writable);
1497}
1498
1499static inline int check_user_page_hwpoison(unsigned long addr)
1500{
1501        int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1502
1503        rc = get_user_pages(addr, 1, flags, NULL, NULL);
1504        return rc == -EHWPOISON;
1505}
1506
1507/*
1508 * The fast path to get the writable pfn which will be stored in @pfn,
1509 * true indicates success, otherwise false is returned.  It's also the
1510 * only part that runs if we can are in atomic context.
1511 */
1512static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
1513                            bool *writable, kvm_pfn_t *pfn)
1514{
1515        struct page *page[1];
1516        int npages;
1517
1518        /*
1519         * Fast pin a writable pfn only if it is a write fault request
1520         * or the caller allows to map a writable pfn for a read fault
1521         * request.
1522         */
1523        if (!(write_fault || writable))
1524                return false;
1525
1526        npages = __get_user_pages_fast(addr, 1, 1, page);
1527        if (npages == 1) {
1528                *pfn = page_to_pfn(page[0]);
1529
1530                if (writable)
1531                        *writable = true;
1532                return true;
1533        }
1534
1535        return false;
1536}
1537
1538/*
1539 * The slow path to get the pfn of the specified host virtual address,
1540 * 1 indicates success, -errno is returned if error is detected.
1541 */
1542static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1543                           bool *writable, kvm_pfn_t *pfn)
1544{
1545        unsigned int flags = FOLL_HWPOISON;
1546        struct page *page;
1547        int npages = 0;
1548
1549        might_sleep();
1550
1551        if (writable)
1552                *writable = write_fault;
1553
1554        if (write_fault)
1555                flags |= FOLL_WRITE;
1556        if (async)
1557                flags |= FOLL_NOWAIT;
1558
1559        npages = get_user_pages_unlocked(addr, 1, &page, flags);
1560        if (npages != 1)
1561                return npages;
1562
1563        /* map read fault as writable if possible */
1564        if (unlikely(!write_fault) && writable) {
1565                struct page *wpage;
1566
1567                if (__get_user_pages_fast(addr, 1, 1, &wpage) == 1) {
1568                        *writable = true;
1569                        put_page(page);
1570                        page = wpage;
1571                }
1572        }
1573        *pfn = page_to_pfn(page);
1574        return npages;
1575}
1576
1577static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1578{
1579        if (unlikely(!(vma->vm_flags & VM_READ)))
1580                return false;
1581
1582        if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1583                return false;
1584
1585        return true;
1586}
1587
1588static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1589                               unsigned long addr, bool *async,
1590                               bool write_fault, bool *writable,
1591                               kvm_pfn_t *p_pfn)
1592{
1593        unsigned long pfn;
1594        int r;
1595
1596        r = follow_pfn(vma, addr, &pfn);
1597        if (r) {
1598                /*
1599                 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1600                 * not call the fault handler, so do it here.
1601                 */
1602                bool unlocked = false;
1603                r = fixup_user_fault(current, current->mm, addr,
1604                                     (write_fault ? FAULT_FLAG_WRITE : 0),
1605                                     &unlocked);
1606                if (unlocked)
1607                        return -EAGAIN;
1608                if (r)
1609                        return r;
1610
1611                r = follow_pfn(vma, addr, &pfn);
1612                if (r)
1613                        return r;
1614
1615        }
1616
1617        if (writable)
1618                *writable = true;
1619
1620        /*
1621         * Get a reference here because callers of *hva_to_pfn* and
1622         * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1623         * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
1624         * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1625         * simply do nothing for reserved pfns.
1626         *
1627         * Whoever called remap_pfn_range is also going to call e.g.
1628         * unmap_mapping_range before the underlying pages are freed,
1629         * causing a call to our MMU notifier.
1630         */ 
1631        kvm_get_pfn(pfn);
1632
1633        *p_pfn = pfn;
1634        return 0;
1635}
1636
1637/*
1638 * Pin guest page in memory and return its pfn.
1639 * @addr: host virtual address which maps memory to the guest
1640 * @atomic: whether this function can sleep
1641 * @async: whether this function need to wait IO complete if the
1642 *         host page is not in the memory
1643 * @write_fault: whether we should get a writable host page
1644 * @writable: whether it allows to map a writable host page for !@write_fault
1645 *
1646 * The function will map a writable host page for these two cases:
1647 * 1): @write_fault = true
1648 * 2): @write_fault = false && @writable, @writable will tell the caller
1649 *     whether the mapping is writable.
1650 */
1651static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1652                        bool write_fault, bool *writable)
1653{
1654        struct vm_area_struct *vma;
1655        kvm_pfn_t pfn = 0;
1656        int npages, r;
1657
1658        /* we can do it either atomically or asynchronously, not both */
1659        BUG_ON(atomic && async);
1660
1661        if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
1662                return pfn;
1663
1664        if (atomic)
1665                return KVM_PFN_ERR_FAULT;
1666
1667        npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1668        if (npages == 1)
1669                return pfn;
1670
1671        down_read(&current->mm->mmap_sem);
1672        if (npages == -EHWPOISON ||
1673              (!async && check_user_page_hwpoison(addr))) {
1674                pfn = KVM_PFN_ERR_HWPOISON;
1675                goto exit;
1676        }
1677
1678retry:
1679        vma = find_vma_intersection(current->mm, addr, addr + 1);
1680
1681        if (vma == NULL)
1682                pfn = KVM_PFN_ERR_FAULT;
1683        else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1684                r = hva_to_pfn_remapped(vma, addr, async, write_fault, writable, &pfn);
1685                if (r == -EAGAIN)
1686                        goto retry;
1687                if (r < 0)
1688                        pfn = KVM_PFN_ERR_FAULT;
1689        } else {
1690                if (async && vma_is_valid(vma, write_fault))
1691                        *async = true;
1692                pfn = KVM_PFN_ERR_FAULT;
1693        }
1694exit:
1695        up_read(&current->mm->mmap_sem);
1696        return pfn;
1697}
1698
1699kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1700                               bool atomic, bool *async, bool write_fault,
1701                               bool *writable)
1702{
1703        unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1704
1705        if (addr == KVM_HVA_ERR_RO_BAD) {
1706                if (writable)
1707                        *writable = false;
1708                return KVM_PFN_ERR_RO_FAULT;
1709        }
1710
1711        if (kvm_is_error_hva(addr)) {
1712                if (writable)
1713                        *writable = false;
1714                return KVM_PFN_NOSLOT;
1715        }
1716
1717        /* Do not map writable pfn in the readonly memslot. */
1718        if (writable && memslot_is_readonly(slot)) {
1719                *writable = false;
1720                writable = NULL;
1721        }
1722
1723        return hva_to_pfn(addr, atomic, async, write_fault,
1724                          writable);
1725}
1726EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1727
1728kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1729                      bool *writable)
1730{
1731        return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1732                                    write_fault, writable);
1733}
1734EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1735
1736kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1737{
1738        return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1739}
1740EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1741
1742kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1743{
1744        return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1745}
1746EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1747
1748kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1749{
1750        return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1751}
1752EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1753
1754kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1755{
1756        return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1757}
1758EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1759
1760kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1761{
1762        return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1763}
1764EXPORT_SYMBOL_GPL(gfn_to_pfn);
1765
1766kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1767{
1768        return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1769}
1770EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1771
1772int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1773                            struct page **pages, int nr_pages)
1774{
1775        unsigned long addr;
1776        gfn_t entry = 0;
1777
1778        addr = gfn_to_hva_many(slot, gfn, &entry);
1779        if (kvm_is_error_hva(addr))
1780                return -1;
1781
1782        if (entry < nr_pages)
1783                return 0;
1784
1785        return __get_user_pages_fast(addr, nr_pages, 1, pages);
1786}
1787EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1788
1789static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1790{
1791        if (is_error_noslot_pfn(pfn))
1792                return KVM_ERR_PTR_BAD_PAGE;
1793
1794        if (kvm_is_reserved_pfn(pfn)) {
1795                WARN_ON(1);
1796                return KVM_ERR_PTR_BAD_PAGE;
1797        }
1798
1799        return pfn_to_page(pfn);
1800}
1801
1802struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1803{
1804        kvm_pfn_t pfn;
1805
1806        pfn = gfn_to_pfn(kvm, gfn);
1807
1808        return kvm_pfn_to_page(pfn);
1809}
1810EXPORT_SYMBOL_GPL(gfn_to_page);
1811
1812static int __kvm_map_gfn(struct kvm_memory_slot *slot, gfn_t gfn,
1813                         struct kvm_host_map *map)
1814{
1815        kvm_pfn_t pfn;
1816        void *hva = NULL;
1817        struct page *page = KVM_UNMAPPED_PAGE;
1818
1819        if (!map)
1820                return -EINVAL;
1821
1822        pfn = gfn_to_pfn_memslot(slot, gfn);
1823        if (is_error_noslot_pfn(pfn))
1824                return -EINVAL;
1825
1826        if (pfn_valid(pfn)) {
1827                page = pfn_to_page(pfn);
1828                hva = kmap(page);
1829#ifdef CONFIG_HAS_IOMEM
1830        } else {
1831                hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
1832#endif
1833        }
1834
1835        if (!hva)
1836                return -EFAULT;
1837
1838        map->page = page;
1839        map->hva = hva;
1840        map->pfn = pfn;
1841        map->gfn = gfn;
1842
1843        return 0;
1844}
1845
1846int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
1847{
1848        return __kvm_map_gfn(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, map);
1849}
1850EXPORT_SYMBOL_GPL(kvm_vcpu_map);
1851
1852void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map,
1853                    bool dirty)
1854{
1855        if (!map)
1856                return;
1857
1858        if (!map->hva)
1859                return;
1860
1861        if (map->page != KVM_UNMAPPED_PAGE)
1862                kunmap(map->page);
1863#ifdef CONFIG_HAS_IOMEM
1864        else
1865                memunmap(map->hva);
1866#endif
1867
1868        if (dirty) {
1869                kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
1870                kvm_release_pfn_dirty(map->pfn);
1871        } else {
1872                kvm_release_pfn_clean(map->pfn);
1873        }
1874
1875        map->hva = NULL;
1876        map->page = NULL;
1877}
1878EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
1879
1880struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1881{
1882        kvm_pfn_t pfn;
1883
1884        pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1885
1886        return kvm_pfn_to_page(pfn);
1887}
1888EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1889
1890void kvm_release_page_clean(struct page *page)
1891{
1892        WARN_ON(is_error_page(page));
1893
1894        kvm_release_pfn_clean(page_to_pfn(page));
1895}
1896EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1897
1898void kvm_release_pfn_clean(kvm_pfn_t pfn)
1899{
1900        if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1901                put_page(pfn_to_page(pfn));
1902}
1903EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1904
1905void kvm_release_page_dirty(struct page *page)
1906{
1907        WARN_ON(is_error_page(page));
1908
1909        kvm_release_pfn_dirty(page_to_pfn(page));
1910}
1911EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1912
1913void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1914{
1915        kvm_set_pfn_dirty(pfn);
1916        kvm_release_pfn_clean(pfn);
1917}
1918EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1919
1920void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1921{
1922        if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn)) {
1923                struct page *page = pfn_to_page(pfn);
1924
1925                SetPageDirty(page);
1926        }
1927}
1928EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1929
1930void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1931{
1932        if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
1933                mark_page_accessed(pfn_to_page(pfn));
1934}
1935EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1936
1937void kvm_get_pfn(kvm_pfn_t pfn)
1938{
1939        if (!kvm_is_reserved_pfn(pfn))
1940                get_page(pfn_to_page(pfn));
1941}
1942EXPORT_SYMBOL_GPL(kvm_get_pfn);
1943
1944static int next_segment(unsigned long len, int offset)
1945{
1946        if (len > PAGE_SIZE - offset)
1947                return PAGE_SIZE - offset;
1948        else
1949                return len;
1950}
1951
1952static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1953                                 void *data, int offset, int len)
1954{
1955        int r;
1956        unsigned long addr;
1957
1958        addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1959        if (kvm_is_error_hva(addr))
1960                return -EFAULT;
1961        r = __copy_from_user(data, (void __user *)addr + offset, len);
1962        if (r)
1963                return -EFAULT;
1964        return 0;
1965}
1966
1967int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1968                        int len)
1969{
1970        struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1971
1972        return __kvm_read_guest_page(slot, gfn, data, offset, len);
1973}
1974EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1975
1976int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1977                             int offset, int len)
1978{
1979        struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1980
1981        return __kvm_read_guest_page(slot, gfn, data, offset, len);
1982}
1983EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1984
1985int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1986{
1987        gfn_t gfn = gpa >> PAGE_SHIFT;
1988        int seg;
1989        int offset = offset_in_page(gpa);
1990        int ret;
1991
1992        while ((seg = next_segment(len, offset)) != 0) {
1993                ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1994                if (ret < 0)
1995                        return ret;
1996                offset = 0;
1997                len -= seg;
1998                data += seg;
1999                ++gfn;
2000        }
2001        return 0;
2002}
2003EXPORT_SYMBOL_GPL(kvm_read_guest);
2004
2005int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2006{
2007        gfn_t gfn = gpa >> PAGE_SHIFT;
2008        int seg;
2009        int offset = offset_in_page(gpa);
2010        int ret;
2011
2012        while ((seg = next_segment(len, offset)) != 0) {
2013                ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2014                if (ret < 0)
2015                        return ret;
2016                offset = 0;
2017                len -= seg;
2018                data += seg;
2019                ++gfn;
2020        }
2021        return 0;
2022}
2023EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2024
2025static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2026                                   void *data, int offset, unsigned long len)
2027{
2028        int r;
2029        unsigned long addr;
2030
2031        addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2032        if (kvm_is_error_hva(addr))
2033                return -EFAULT;
2034        pagefault_disable();
2035        r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2036        pagefault_enable();
2037        if (r)
2038                return -EFAULT;
2039        return 0;
2040}
2041
2042int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
2043                          unsigned long len)
2044{
2045        gfn_t gfn = gpa >> PAGE_SHIFT;
2046        struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2047        int offset = offset_in_page(gpa);
2048
2049        return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2050}
2051EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
2052
2053int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2054                               void *data, unsigned long len)
2055{
2056        gfn_t gfn = gpa >> PAGE_SHIFT;
2057        struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2058        int offset = offset_in_page(gpa);
2059
2060        return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2061}
2062EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2063
2064static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
2065                                  const void *data, int offset, int len)
2066{
2067        int r;
2068        unsigned long addr;
2069
2070        addr = gfn_to_hva_memslot(memslot, gfn);
2071        if (kvm_is_error_hva(addr))
2072                return -EFAULT;
2073        r = __copy_to_user((void __user *)addr + offset, data, len);
2074        if (r)
2075                return -EFAULT;
2076        mark_page_dirty_in_slot(memslot, gfn);
2077        return 0;
2078}
2079
2080int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2081                         const void *data, int offset, int len)
2082{
2083        struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2084
2085        return __kvm_write_guest_page(slot, gfn, data, offset, len);
2086}
2087EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2088
2089int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2090                              const void *data, int offset, int len)
2091{
2092        struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2093
2094        return __kvm_write_guest_page(slot, gfn, data, offset, len);
2095}
2096EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2097
2098int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2099                    unsigned long len)
2100{
2101        gfn_t gfn = gpa >> PAGE_SHIFT;
2102        int seg;
2103        int offset = offset_in_page(gpa);
2104        int ret;
2105
2106        while ((seg = next_segment(len, offset)) != 0) {
2107                ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2108                if (ret < 0)
2109                        return ret;
2110                offset = 0;
2111                len -= seg;
2112                data += seg;
2113                ++gfn;
2114        }
2115        return 0;
2116}
2117EXPORT_SYMBOL_GPL(kvm_write_guest);
2118
2119int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2120                         unsigned long len)
2121{
2122        gfn_t gfn = gpa >> PAGE_SHIFT;
2123        int seg;
2124        int offset = offset_in_page(gpa);
2125        int ret;
2126
2127        while ((seg = next_segment(len, offset)) != 0) {
2128                ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
2129                if (ret < 0)
2130                        return ret;
2131                offset = 0;
2132                len -= seg;
2133                data += seg;
2134                ++gfn;
2135        }
2136        return 0;
2137}
2138EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
2139
2140static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
2141                                       struct gfn_to_hva_cache *ghc,
2142                                       gpa_t gpa, unsigned long len)
2143{
2144        int offset = offset_in_page(gpa);
2145        gfn_t start_gfn = gpa >> PAGE_SHIFT;
2146        gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
2147        gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
2148        gfn_t nr_pages_avail;
2149        int r = start_gfn <= end_gfn ? 0 : -EINVAL;
2150
2151        ghc->gpa = gpa;
2152        ghc->generation = slots->generation;
2153        ghc->len = len;
2154        ghc->hva = KVM_HVA_ERR_BAD;
2155
2156        /*
2157         * If the requested region crosses two memslots, we still
2158         * verify that the entire region is valid here.
2159         */
2160        while (!r && start_gfn <= end_gfn) {
2161                ghc->memslot = __gfn_to_memslot(slots, start_gfn);
2162                ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
2163                                           &nr_pages_avail);
2164                if (kvm_is_error_hva(ghc->hva))
2165                        r = -EFAULT;
2166                start_gfn += nr_pages_avail;
2167        }
2168
2169        /* Use the slow path for cross page reads and writes. */
2170        if (!r && nr_pages_needed == 1)
2171                ghc->hva += offset;
2172        else
2173                ghc->memslot = NULL;
2174
2175        return r;
2176}
2177
2178int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2179                              gpa_t gpa, unsigned long len)
2180{
2181        struct kvm_memslots *slots = kvm_memslots(kvm);
2182        return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
2183}
2184EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
2185
2186int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2187                                  void *data, unsigned int offset,
2188                                  unsigned long len)
2189{
2190        struct kvm_memslots *slots = kvm_memslots(kvm);
2191        int r;
2192        gpa_t gpa = ghc->gpa + offset;
2193
2194        BUG_ON(len + offset > ghc->len);
2195
2196        if (slots->generation != ghc->generation)
2197                __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2198
2199        if (unlikely(!ghc->memslot))
2200                return kvm_write_guest(kvm, gpa, data, len);
2201
2202        if (kvm_is_error_hva(ghc->hva))
2203                return -EFAULT;
2204
2205        r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2206        if (r)
2207                return -EFAULT;
2208        mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2209
2210        return 0;
2211}
2212EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2213
2214int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2215                           void *data, unsigned long len)
2216{
2217        return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2218}
2219EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2220
2221int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2222                           void *data, unsigned long len)
2223{
2224        struct kvm_memslots *slots = kvm_memslots(kvm);
2225        int r;
2226
2227        BUG_ON(len > ghc->len);
2228
2229        if (slots->generation != ghc->generation)
2230                __kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2231
2232        if (unlikely(!ghc->memslot))
2233                return kvm_read_guest(kvm, ghc->gpa, data, len);
2234
2235        if (kvm_is_error_hva(ghc->hva))
2236                return -EFAULT;
2237
2238        r = __copy_from_user(data, (void __user *)ghc->hva, len);
2239        if (r)
2240                return -EFAULT;
2241
2242        return 0;
2243}
2244EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2245
2246int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2247{
2248        const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2249
2250        return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2251}
2252EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2253
2254int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2255{
2256        gfn_t gfn = gpa >> PAGE_SHIFT;
2257        int seg;
2258        int offset = offset_in_page(gpa);
2259        int ret;
2260
2261        while ((seg = next_segment(len, offset)) != 0) {
2262                ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2263                if (ret < 0)
2264                        return ret;
2265                offset = 0;
2266                len -= seg;
2267                ++gfn;
2268        }
2269        return 0;
2270}
2271EXPORT_SYMBOL_GPL(kvm_clear_guest);
2272
2273static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2274                                    gfn_t gfn)
2275{
2276        if (memslot && memslot->dirty_bitmap) {
2277                unsigned long rel_gfn = gfn - memslot->base_gfn;
2278
2279                set_bit_le(rel_gfn, memslot->dirty_bitmap);
2280        }
2281}
2282
2283void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2284{
2285        struct kvm_memory_slot *memslot;
2286
2287        memslot = gfn_to_memslot(kvm, gfn);
2288        mark_page_dirty_in_slot(memslot, gfn);
2289}
2290EXPORT_SYMBOL_GPL(mark_page_dirty);
2291
2292void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2293{
2294        struct kvm_memory_slot *memslot;
2295
2296        memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2297        mark_page_dirty_in_slot(memslot, gfn);
2298}
2299EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2300
2301void kvm_sigset_activate(struct kvm_vcpu *vcpu)
2302{
2303        if (!vcpu->sigset_active)
2304                return;
2305
2306        /*
2307         * This does a lockless modification of ->real_blocked, which is fine
2308         * because, only current can change ->real_blocked and all readers of
2309         * ->real_blocked don't care as long ->real_blocked is always a subset
2310         * of ->blocked.
2311         */
2312        sigprocmask(SIG_SETMASK, &vcpu->sigset, &current->real_blocked);
2313}
2314
2315void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
2316{
2317        if (!vcpu->sigset_active)
2318                return;
2319
2320        sigprocmask(SIG_SETMASK, &current->real_blocked, NULL);
2321        sigemptyset(&current->real_blocked);
2322}
2323
2324static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2325{
2326        unsigned int old, val, grow, grow_start;
2327
2328        old = val = vcpu->halt_poll_ns;
2329        grow_start = READ_ONCE(halt_poll_ns_grow_start);
2330        grow = READ_ONCE(halt_poll_ns_grow);
2331        if (!grow)
2332                goto out;
2333
2334        val *= grow;
2335        if (val < grow_start)
2336                val = grow_start;
2337
2338        if (val > halt_poll_ns)
2339                val = halt_poll_ns;
2340
2341        vcpu->halt_poll_ns = val;
2342out:
2343        trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2344}
2345
2346static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2347{
2348        unsigned int old, val, shrink;
2349
2350        old = val = vcpu->halt_poll_ns;
2351        shrink = READ_ONCE(halt_poll_ns_shrink);
2352        if (shrink == 0)
2353                val = 0;
2354        else
2355                val /= shrink;
2356
2357        vcpu->halt_poll_ns = val;
2358        trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2359}
2360
2361static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2362{
2363        int ret = -EINTR;
2364        int idx = srcu_read_lock(&vcpu->kvm->srcu);
2365
2366        if (kvm_arch_vcpu_runnable(vcpu)) {
2367                kvm_make_request(KVM_REQ_UNHALT, vcpu);
2368                goto out;
2369        }
2370        if (kvm_cpu_has_pending_timer(vcpu))
2371                goto out;
2372        if (signal_pending(current))
2373                goto out;
2374
2375        ret = 0;
2376out:
2377        srcu_read_unlock(&vcpu->kvm->srcu, idx);
2378        return ret;
2379}
2380
2381/*
2382 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2383 */
2384void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2385{
2386        ktime_t start, cur;
2387        DECLARE_SWAITQUEUE(wait);
2388        bool waited = false;
2389        u64 block_ns;
2390
2391        kvm_arch_vcpu_blocking(vcpu);
2392
2393        start = cur = ktime_get();
2394        if (vcpu->halt_poll_ns && !kvm_arch_no_poll(vcpu)) {
2395                ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2396
2397                ++vcpu->stat.halt_attempted_poll;
2398                do {
2399                        /*
2400                         * This sets KVM_REQ_UNHALT if an interrupt
2401                         * arrives.
2402                         */
2403                        if (kvm_vcpu_check_block(vcpu) < 0) {
2404                                ++vcpu->stat.halt_successful_poll;
2405                                if (!vcpu_valid_wakeup(vcpu))
2406                                        ++vcpu->stat.halt_poll_invalid;
2407                                goto out;
2408                        }
2409                        cur = ktime_get();
2410                } while (single_task_running() && ktime_before(cur, stop));
2411        }
2412
2413        for (;;) {
2414                prepare_to_swait_exclusive(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2415
2416                if (kvm_vcpu_check_block(vcpu) < 0)
2417                        break;
2418
2419                waited = true;
2420                schedule();
2421        }
2422
2423        finish_swait(&vcpu->wq, &wait);
2424        cur = ktime_get();
2425out:
2426        kvm_arch_vcpu_unblocking(vcpu);
2427        block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2428
2429        if (!kvm_arch_no_poll(vcpu)) {
2430                if (!vcpu_valid_wakeup(vcpu)) {
2431                        shrink_halt_poll_ns(vcpu);
2432                } else if (halt_poll_ns) {
2433                        if (block_ns <= vcpu->halt_poll_ns)
2434                                ;
2435                        /* we had a long block, shrink polling */
2436                        else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2437                                shrink_halt_poll_ns(vcpu);
2438                        /* we had a short halt and our poll time is too small */
2439                        else if (vcpu->halt_poll_ns < halt_poll_ns &&
2440                                block_ns < halt_poll_ns)
2441                                grow_halt_poll_ns(vcpu);
2442                } else {
2443                        vcpu->halt_poll_ns = 0;
2444                }
2445        }
2446
2447        trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2448        kvm_arch_vcpu_block_finish(vcpu);
2449}
2450EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2451
2452bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2453{
2454        struct swait_queue_head *wqp;
2455
2456        wqp = kvm_arch_vcpu_wq(vcpu);
2457        if (swq_has_sleeper(wqp)) {
2458                swake_up_one(wqp);
2459                WRITE_ONCE(vcpu->ready, true);
2460                ++vcpu->stat.halt_wakeup;
2461                return true;
2462        }
2463
2464        return false;
2465}
2466EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2467
2468#ifndef CONFIG_S390
2469/*
2470 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2471 */
2472void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2473{
2474        int me;
2475        int cpu = vcpu->cpu;
2476
2477        if (kvm_vcpu_wake_up(vcpu))
2478                return;
2479
2480        me = get_cpu();
2481        if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2482                if (kvm_arch_vcpu_should_kick(vcpu))
2483                        smp_send_reschedule(cpu);
2484        put_cpu();
2485}
2486EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2487#endif /* !CONFIG_S390 */
2488
2489int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2490{
2491        struct pid *pid;
2492        struct task_struct *task = NULL;
2493        int ret = 0;
2494
2495        rcu_read_lock();
2496        pid = rcu_dereference(target->pid);
2497        if (pid)
2498                task = get_pid_task(pid, PIDTYPE_PID);
2499        rcu_read_unlock();
2500        if (!task)
2501                return ret;
2502        ret = yield_to(task, 1);
2503        put_task_struct(task);
2504
2505        return ret;
2506}
2507EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2508
2509/*
2510 * Helper that checks whether a VCPU is eligible for directed yield.
2511 * Most eligible candidate to yield is decided by following heuristics:
2512 *
2513 *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2514 *  (preempted lock holder), indicated by @in_spin_loop.
2515 *  Set at the beiginning and cleared at the end of interception/PLE handler.
2516 *
2517 *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2518 *  chance last time (mostly it has become eligible now since we have probably
2519 *  yielded to lockholder in last iteration. This is done by toggling
2520 *  @dy_eligible each time a VCPU checked for eligibility.)
2521 *
2522 *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2523 *  to preempted lock-holder could result in wrong VCPU selection and CPU
2524 *  burning. Giving priority for a potential lock-holder increases lock
2525 *  progress.
2526 *
2527 *  Since algorithm is based on heuristics, accessing another VCPU data without
2528 *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2529 *  and continue with next VCPU and so on.
2530 */
2531static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2532{
2533#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2534        bool eligible;
2535
2536        eligible = !vcpu->spin_loop.in_spin_loop ||
2537                    vcpu->spin_loop.dy_eligible;
2538
2539        if (vcpu->spin_loop.in_spin_loop)
2540                kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2541
2542        return eligible;
2543#else
2544        return true;
2545#endif
2546}
2547
2548/*
2549 * Unlike kvm_arch_vcpu_runnable, this function is called outside
2550 * a vcpu_load/vcpu_put pair.  However, for most architectures
2551 * kvm_arch_vcpu_runnable does not require vcpu_load.
2552 */
2553bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
2554{
2555        return kvm_arch_vcpu_runnable(vcpu);
2556}
2557
2558static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
2559{
2560        if (kvm_arch_dy_runnable(vcpu))
2561                return true;
2562
2563#ifdef CONFIG_KVM_ASYNC_PF
2564        if (!list_empty_careful(&vcpu->async_pf.done))
2565                return true;
2566#endif
2567
2568        return false;
2569}
2570
2571void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
2572{
2573        struct kvm *kvm = me->kvm;
2574        struct kvm_vcpu *vcpu;
2575        int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2576        int yielded = 0;
2577        int try = 3;
2578        int pass;
2579        int i;
2580
2581        kvm_vcpu_set_in_spin_loop(me, true);
2582        /*
2583         * We boost the priority of a VCPU that is runnable but not
2584         * currently running, because it got preempted by something
2585         * else and called schedule in __vcpu_run.  Hopefully that
2586         * VCPU is holding the lock that we need and will release it.
2587         * We approximate round-robin by starting at the last boosted VCPU.
2588         */
2589        for (pass = 0; pass < 2 && !yielded && try; pass++) {
2590                kvm_for_each_vcpu(i, vcpu, kvm) {
2591                        if (!pass && i <= last_boosted_vcpu) {
2592                                i = last_boosted_vcpu;
2593                                continue;
2594                        } else if (pass && i > last_boosted_vcpu)
2595                                break;
2596                        if (!READ_ONCE(vcpu->ready))
2597                                continue;
2598                        if (vcpu == me)
2599                                continue;
2600                        if (swait_active(&vcpu->wq) && !vcpu_dy_runnable(vcpu))
2601                                continue;
2602                        if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
2603                                !kvm_arch_vcpu_in_kernel(vcpu))
2604                                continue;
2605                        if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2606                                continue;
2607
2608                        yielded = kvm_vcpu_yield_to(vcpu);
2609                        if (yielded > 0) {
2610                                kvm->last_boosted_vcpu = i;
2611                                break;
2612                        } else if (yielded < 0) {
2613                                try--;
2614                                if (!try)
2615                                        break;
2616                        }
2617                }
2618        }
2619        kvm_vcpu_set_in_spin_loop(me, false);
2620
2621        /* Ensure vcpu is not eligible during next spinloop */
2622        kvm_vcpu_set_dy_eligible(me, false);
2623}
2624EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2625
2626static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
2627{
2628        struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2629        struct page *page;
2630
2631        if (vmf->pgoff == 0)
2632                page = virt_to_page(vcpu->run);
2633#ifdef CONFIG_X86
2634        else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2635                page = virt_to_page(vcpu->arch.pio_data);
2636#endif
2637#ifdef CONFIG_KVM_MMIO
2638        else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2639                page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2640#endif
2641        else
2642                return kvm_arch_vcpu_fault(vcpu, vmf);
2643        get_page(page);
2644        vmf->page = page;
2645        return 0;
2646}
2647
2648static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2649        .fault = kvm_vcpu_fault,
2650};
2651
2652static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2653{
2654        vma->vm_ops = &kvm_vcpu_vm_ops;
2655        return 0;
2656}
2657
2658static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2659{
2660        struct kvm_vcpu *vcpu = filp->private_data;
2661
2662        debugfs_remove_recursive(vcpu->debugfs_dentry);
2663        kvm_put_kvm(vcpu->kvm);
2664        return 0;
2665}
2666
2667static struct file_operations kvm_vcpu_fops = {
2668        .release        = kvm_vcpu_release,
2669        .unlocked_ioctl = kvm_vcpu_ioctl,
2670        .mmap           = kvm_vcpu_mmap,
2671        .llseek         = noop_llseek,
2672        KVM_COMPAT(kvm_vcpu_compat_ioctl),
2673};
2674
2675/*
2676 * Allocates an inode for the vcpu.
2677 */
2678static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2679{
2680        char name[8 + 1 + ITOA_MAX_LEN + 1];
2681
2682        snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
2683        return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2684}
2685
2686static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2687{
2688#ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
2689        char dir_name[ITOA_MAX_LEN * 2];
2690
2691        if (!debugfs_initialized())
2692                return;
2693
2694        snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2695        vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2696                                                  vcpu->kvm->debugfs_dentry);
2697
2698        kvm_arch_create_vcpu_debugfs(vcpu);
2699#endif
2700}
2701
2702/*
2703 * Creates some virtual cpus.  Good luck creating more than one.
2704 */
2705static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2706{
2707        int r;
2708        struct kvm_vcpu *vcpu;
2709
2710        if (id >= KVM_MAX_VCPU_ID)
2711                return -EINVAL;
2712
2713        mutex_lock(&kvm->lock);
2714        if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2715                mutex_unlock(&kvm->lock);
2716                return -EINVAL;
2717        }
2718
2719        kvm->created_vcpus++;
2720        mutex_unlock(&kvm->lock);
2721
2722        vcpu = kvm_arch_vcpu_create(kvm, id);
2723        if (IS_ERR(vcpu)) {
2724                r = PTR_ERR(vcpu);
2725                goto vcpu_decrement;
2726        }
2727
2728        preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2729
2730        r = kvm_arch_vcpu_setup(vcpu);
2731        if (r)
2732                goto vcpu_destroy;
2733
2734        kvm_create_vcpu_debugfs(vcpu);
2735
2736        mutex_lock(&kvm->lock);
2737        if (kvm_get_vcpu_by_id(kvm, id)) {
2738                r = -EEXIST;
2739                goto unlock_vcpu_destroy;
2740        }
2741
2742        BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2743
2744        /* Now it's all set up, let userspace reach it */
2745        kvm_get_kvm(kvm);
2746        r = create_vcpu_fd(vcpu);
2747        if (r < 0) {
2748                kvm_put_kvm(kvm);
2749                goto unlock_vcpu_destroy;
2750        }
2751
2752        kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2753
2754        /*
2755         * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2756         * before kvm->online_vcpu's incremented value.
2757         */
2758        smp_wmb();
2759        atomic_inc(&kvm->online_vcpus);
2760
2761        mutex_unlock(&kvm->lock);
2762        kvm_arch_vcpu_postcreate(vcpu);
2763        return r;
2764
2765unlock_vcpu_destroy:
2766        mutex_unlock(&kvm->lock);
2767        debugfs_remove_recursive(vcpu->debugfs_dentry);
2768vcpu_destroy:
2769        kvm_arch_vcpu_destroy(vcpu);
2770vcpu_decrement:
2771        mutex_lock(&kvm->lock);
2772        kvm->created_vcpus--;
2773        mutex_unlock(&kvm->lock);
2774        return r;
2775}
2776
2777static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2778{
2779        if (sigset) {
2780                sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2781                vcpu->sigset_active = 1;
2782                vcpu->sigset = *sigset;
2783        } else
2784                vcpu->sigset_active = 0;
2785        return 0;
2786}
2787
2788static long kvm_vcpu_ioctl(struct file *filp,
2789                           unsigned int ioctl, unsigned long arg)
2790{
2791        struct kvm_vcpu *vcpu = filp->private_data;
2792        void __user *argp = (void __user *)arg;
2793        int r;
2794        struct kvm_fpu *fpu = NULL;
2795        struct kvm_sregs *kvm_sregs = NULL;
2796
2797        if (vcpu->kvm->mm != current->mm)
2798                return -EIO;
2799
2800        if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2801                return -EINVAL;
2802
2803        /*
2804         * Some architectures have vcpu ioctls that are asynchronous to vcpu
2805         * execution; mutex_lock() would break them.
2806         */
2807        r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
2808        if (r != -ENOIOCTLCMD)
2809                return r;
2810
2811        if (mutex_lock_killable(&vcpu->mutex))
2812                return -EINTR;
2813        switch (ioctl) {
2814        case KVM_RUN: {
2815                struct pid *oldpid;
2816                r = -EINVAL;
2817                if (arg)
2818                        goto out;
2819                oldpid = rcu_access_pointer(vcpu->pid);
2820                if (unlikely(oldpid != task_pid(current))) {
2821                        /* The thread running this VCPU changed. */
2822                        struct pid *newpid;
2823
2824                        r = kvm_arch_vcpu_run_pid_change(vcpu);
2825                        if (r)
2826                                break;
2827
2828                        newpid = get_task_pid(current, PIDTYPE_PID);
2829                        rcu_assign_pointer(vcpu->pid, newpid);
2830                        if (oldpid)
2831                                synchronize_rcu();
2832                        put_pid(oldpid);
2833                }
2834                r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2835                trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2836                break;
2837        }
2838        case KVM_GET_REGS: {
2839                struct kvm_regs *kvm_regs;
2840
2841                r = -ENOMEM;
2842                kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
2843                if (!kvm_regs)
2844                        goto out;
2845                r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2846                if (r)
2847                        goto out_free1;
2848                r = -EFAULT;
2849                if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2850                        goto out_free1;
2851                r = 0;
2852out_free1:
2853                kfree(kvm_regs);
2854                break;
2855        }
2856        case KVM_SET_REGS: {
2857                struct kvm_regs *kvm_regs;
2858
2859                r = -ENOMEM;
2860                kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2861                if (IS_ERR(kvm_regs)) {
2862                        r = PTR_ERR(kvm_regs);
2863                        goto out;
2864                }
2865                r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2866                kfree(kvm_regs);
2867                break;
2868        }
2869        case KVM_GET_SREGS: {
2870                kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
2871                                    GFP_KERNEL_ACCOUNT);
2872                r = -ENOMEM;
2873                if (!kvm_sregs)
2874                        goto out;
2875                r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2876                if (r)
2877                        goto out;
2878                r = -EFAULT;
2879                if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2880                        goto out;
2881                r = 0;
2882                break;
2883        }
2884        case KVM_SET_SREGS: {
2885                kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2886                if (IS_ERR(kvm_sregs)) {
2887                        r = PTR_ERR(kvm_sregs);
2888                        kvm_sregs = NULL;
2889                        goto out;
2890                }
2891                r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2892                break;
2893        }
2894        case KVM_GET_MP_STATE: {
2895                struct kvm_mp_state mp_state;
2896
2897                r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2898                if (r)
2899                        goto out;
2900                r = -EFAULT;
2901                if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2902                        goto out;
2903                r = 0;
2904                break;
2905        }
2906        case KVM_SET_MP_STATE: {
2907                struct kvm_mp_state mp_state;
2908
2909                r = -EFAULT;
2910                if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2911                        goto out;
2912                r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2913                break;
2914        }
2915        case KVM_TRANSLATE: {
2916                struct kvm_translation tr;
2917
2918                r = -EFAULT;
2919                if (copy_from_user(&tr, argp, sizeof(tr)))
2920                        goto out;
2921                r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2922                if (r)
2923                        goto out;
2924                r = -EFAULT;
2925                if (copy_to_user(argp, &tr, sizeof(tr)))
2926                        goto out;
2927                r = 0;
2928                break;
2929        }
2930        case KVM_SET_GUEST_DEBUG: {
2931                struct kvm_guest_debug dbg;
2932
2933                r = -EFAULT;
2934                if (copy_from_user(&dbg, argp, sizeof(dbg)))
2935                        goto out;
2936                r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2937                break;
2938        }
2939        case KVM_SET_SIGNAL_MASK: {
2940                struct kvm_signal_mask __user *sigmask_arg = argp;
2941                struct kvm_signal_mask kvm_sigmask;
2942                sigset_t sigset, *p;
2943
2944                p = NULL;
2945                if (argp) {
2946                        r = -EFAULT;
2947                        if (copy_from_user(&kvm_sigmask, argp,
2948                                           sizeof(kvm_sigmask)))
2949                                goto out;
2950                        r = -EINVAL;
2951                        if (kvm_sigmask.len != sizeof(sigset))
2952                                goto out;
2953                        r = -EFAULT;
2954                        if (copy_from_user(&sigset, sigmask_arg->sigset,
2955                                           sizeof(sigset)))
2956                                goto out;
2957                        p = &sigset;
2958                }
2959                r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2960                break;
2961        }
2962        case KVM_GET_FPU: {
2963                fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
2964                r = -ENOMEM;
2965                if (!fpu)
2966                        goto out;
2967                r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2968                if (r)
2969                        goto out;
2970                r = -EFAULT;
2971                if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2972                        goto out;
2973                r = 0;
2974                break;
2975        }
2976        case KVM_SET_FPU: {
2977                fpu = memdup_user(argp, sizeof(*fpu));
2978                if (IS_ERR(fpu)) {
2979                        r = PTR_ERR(fpu);
2980                        fpu = NULL;
2981                        goto out;
2982                }
2983                r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2984                break;
2985        }
2986        default:
2987                r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2988        }
2989out:
2990        mutex_unlock(&vcpu->mutex);
2991        kfree(fpu);
2992        kfree(kvm_sregs);
2993        return r;
2994}
2995
2996#ifdef CONFIG_KVM_COMPAT
2997static long kvm_vcpu_compat_ioctl(struct file *filp,
2998                                  unsigned int ioctl, unsigned long arg)
2999{
3000        struct kvm_vcpu *vcpu = filp->private_data;
3001        void __user *argp = compat_ptr(arg);
3002        int r;
3003
3004        if (vcpu->kvm->mm != current->mm)
3005                return -EIO;
3006
3007        switch (ioctl) {
3008        case KVM_SET_SIGNAL_MASK: {
3009                struct kvm_signal_mask __user *sigmask_arg = argp;
3010                struct kvm_signal_mask kvm_sigmask;
3011                sigset_t sigset;
3012
3013                if (argp) {
3014                        r = -EFAULT;
3015                        if (copy_from_user(&kvm_sigmask, argp,
3016                                           sizeof(kvm_sigmask)))
3017                                goto out;
3018                        r = -EINVAL;
3019                        if (kvm_sigmask.len != sizeof(compat_sigset_t))
3020                                goto out;
3021                        r = -EFAULT;
3022                        if (get_compat_sigset(&sigset, (void *)sigmask_arg->sigset))
3023                                goto out;
3024                        r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
3025                } else
3026                        r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
3027                break;
3028        }
3029        default:
3030                r = kvm_vcpu_ioctl(filp, ioctl, arg);
3031        }
3032
3033out:
3034        return r;
3035}
3036#endif
3037
3038static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
3039{
3040        struct kvm_device *dev = filp->private_data;
3041
3042        if (dev->ops->mmap)
3043                return dev->ops->mmap(dev, vma);
3044
3045        return -ENODEV;
3046}
3047
3048static int kvm_device_ioctl_attr(struct kvm_device *dev,
3049                                 int (*accessor)(struct kvm_device *dev,
3050                                                 struct kvm_device_attr *attr),
3051                                 unsigned long arg)
3052{
3053        struct kvm_device_attr attr;
3054
3055        if (!accessor)
3056                return -EPERM;
3057
3058        if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
3059                return -EFAULT;
3060
3061        return accessor(dev, &attr);
3062}
3063
3064static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
3065                             unsigned long arg)
3066{
3067        struct kvm_device *dev = filp->private_data;
3068
3069        if (dev->kvm->mm != current->mm)
3070                return -EIO;
3071
3072        switch (ioctl) {
3073        case KVM_SET_DEVICE_ATTR:
3074                return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
3075        case KVM_GET_DEVICE_ATTR:
3076                return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
3077        case KVM_HAS_DEVICE_ATTR:
3078                return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
3079        default:
3080                if (dev->ops->ioctl)
3081                        return dev->ops->ioctl(dev, ioctl, arg);
3082
3083                return -ENOTTY;
3084        }
3085}
3086
3087static int kvm_device_release(struct inode *inode, struct file *filp)
3088{
3089        struct kvm_device *dev = filp->private_data;
3090        struct kvm *kvm = dev->kvm;
3091
3092        if (dev->ops->release) {
3093                mutex_lock(&kvm->lock);
3094                list_del(&dev->vm_node);
3095                dev->ops->release(dev);
3096                mutex_unlock(&kvm->lock);
3097        }
3098
3099        kvm_put_kvm(kvm);
3100        return 0;
3101}
3102
3103static const struct file_operations kvm_device_fops = {
3104        .unlocked_ioctl = kvm_device_ioctl,
3105        .release = kvm_device_release,
3106        KVM_COMPAT(kvm_device_ioctl),
3107        .mmap = kvm_device_mmap,
3108};
3109
3110struct kvm_device *kvm_device_from_filp(struct file *filp)
3111{
3112        if (filp->f_op != &kvm_device_fops)
3113                return NULL;
3114
3115        return filp->private_data;
3116}
3117
3118static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
3119#ifdef CONFIG_KVM_MPIC
3120        [KVM_DEV_TYPE_FSL_MPIC_20]      = &kvm_mpic_ops,
3121        [KVM_DEV_TYPE_FSL_MPIC_42]      = &kvm_mpic_ops,
3122#endif
3123};
3124
3125int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
3126{
3127        if (type >= ARRAY_SIZE(kvm_device_ops_table))
3128                return -ENOSPC;
3129
3130        if (kvm_device_ops_table[type] != NULL)
3131                return -EEXIST;
3132
3133        kvm_device_ops_table[type] = ops;
3134        return 0;
3135}
3136
3137void kvm_unregister_device_ops(u32 type)
3138{
3139        if (kvm_device_ops_table[type] != NULL)
3140                kvm_device_ops_table[type] = NULL;
3141}
3142
3143static int kvm_ioctl_create_device(struct kvm *kvm,
3144                                   struct kvm_create_device *cd)
3145{
3146        struct kvm_device_ops *ops = NULL;
3147        struct kvm_device *dev;
3148        bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
3149        int type;
3150        int ret;
3151
3152        if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
3153                return -ENODEV;
3154
3155        type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
3156        ops = kvm_device_ops_table[type];
3157        if (ops == NULL)
3158                return -ENODEV;
3159
3160        if (test)
3161                return 0;
3162
3163        dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
3164        if (!dev)
3165                return -ENOMEM;
3166
3167        dev->ops = ops;
3168        dev->kvm = kvm;
3169
3170        mutex_lock(&kvm->lock);
3171        ret = ops->create(dev, type);
3172        if (ret < 0) {
3173                mutex_unlock(&kvm->lock);
3174                kfree(dev);
3175                return ret;
3176        }
3177        list_add(&dev->vm_node, &kvm->devices);
3178        mutex_unlock(&kvm->lock);
3179
3180        if (ops->init)
3181                ops->init(dev);
3182
3183        kvm_get_kvm(kvm);
3184        ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
3185        if (ret < 0) {
3186                kvm_put_kvm(kvm);
3187                mutex_lock(&kvm->lock);
3188                list_del(&dev->vm_node);
3189                mutex_unlock(&kvm->lock);
3190                ops->destroy(dev);
3191                return ret;
3192        }
3193
3194        cd->fd = ret;
3195        return 0;
3196}
3197
3198static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
3199{
3200        switch (arg) {
3201        case KVM_CAP_USER_MEMORY:
3202        case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
3203        case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
3204        case KVM_CAP_INTERNAL_ERROR_DATA:
3205#ifdef CONFIG_HAVE_KVM_MSI
3206        case KVM_CAP_SIGNAL_MSI:
3207#endif
3208#ifdef CONFIG_HAVE_KVM_IRQFD
3209        case KVM_CAP_IRQFD:
3210        case KVM_CAP_IRQFD_RESAMPLE:
3211#endif
3212        case KVM_CAP_IOEVENTFD_ANY_LENGTH:
3213        case KVM_CAP_CHECK_EXTENSION_VM:
3214        case KVM_CAP_ENABLE_CAP_VM:
3215#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3216        case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3217#endif
3218                return 1;
3219#ifdef CONFIG_KVM_MMIO
3220        case KVM_CAP_COALESCED_MMIO:
3221                return KVM_COALESCED_MMIO_PAGE_OFFSET;
3222        case KVM_CAP_COALESCED_PIO:
3223                return 1;
3224#endif
3225#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3226        case KVM_CAP_IRQ_ROUTING:
3227                return KVM_MAX_IRQ_ROUTES;
3228#endif
3229#if KVM_ADDRESS_SPACE_NUM > 1
3230        case KVM_CAP_MULTI_ADDRESS_SPACE:
3231                return KVM_ADDRESS_SPACE_NUM;
3232#endif
3233        case KVM_CAP_NR_MEMSLOTS:
3234                return KVM_USER_MEM_SLOTS;
3235        default:
3236                break;
3237        }
3238        return kvm_vm_ioctl_check_extension(kvm, arg);
3239}
3240
3241int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
3242                                                  struct kvm_enable_cap *cap)
3243{
3244        return -EINVAL;
3245}
3246
3247static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
3248                                           struct kvm_enable_cap *cap)
3249{
3250        switch (cap->cap) {
3251#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3252        case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
3253                if (cap->flags || (cap->args[0] & ~1))
3254                        return -EINVAL;
3255                kvm->manual_dirty_log_protect = cap->args[0];
3256                return 0;
3257#endif
3258        default:
3259                return kvm_vm_ioctl_enable_cap(kvm, cap);
3260        }
3261}
3262
3263static long kvm_vm_ioctl(struct file *filp,
3264                           unsigned int ioctl, unsigned long arg)
3265{
3266        struct kvm *kvm = filp->private_data;
3267        void __user *argp = (void __user *)arg;
3268        int r;
3269
3270        if (kvm->mm != current->mm)
3271                return -EIO;
3272        switch (ioctl) {
3273        case KVM_CREATE_VCPU:
3274                r = kvm_vm_ioctl_create_vcpu(kvm, arg);
3275                break;
3276        case KVM_ENABLE_CAP: {
3277                struct kvm_enable_cap cap;
3278
3279                r = -EFAULT;
3280                if (copy_from_user(&cap, argp, sizeof(cap)))
3281                        goto out;
3282                r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
3283                break;
3284        }
3285        case KVM_SET_USER_MEMORY_REGION: {
3286                struct kvm_userspace_memory_region kvm_userspace_mem;
3287
3288                r = -EFAULT;
3289                if (copy_from_user(&kvm_userspace_mem, argp,
3290                                                sizeof(kvm_userspace_mem)))
3291                        goto out;
3292
3293                r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
3294                break;
3295        }
3296        case KVM_GET_DIRTY_LOG: {
3297                struct kvm_dirty_log log;
3298
3299                r = -EFAULT;
3300                if (copy_from_user(&log, argp, sizeof(log)))
3301                        goto out;
3302                r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3303                break;
3304        }
3305#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
3306        case KVM_CLEAR_DIRTY_LOG: {
3307                struct kvm_clear_dirty_log log;
3308
3309                r = -EFAULT;
3310                if (copy_from_user(&log, argp, sizeof(log)))
3311                        goto out;
3312                r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
3313                break;
3314        }
3315#endif
3316#ifdef CONFIG_KVM_MMIO
3317        case KVM_REGISTER_COALESCED_MMIO: {
3318                struct kvm_coalesced_mmio_zone zone;
3319
3320                r = -EFAULT;
3321                if (copy_from_user(&zone, argp, sizeof(zone)))
3322                        goto out;
3323                r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3324                break;
3325        }
3326        case KVM_UNREGISTER_COALESCED_MMIO: {
3327                struct kvm_coalesced_mmio_zone zone;
3328
3329                r = -EFAULT;
3330                if (copy_from_user(&zone, argp, sizeof(zone)))
3331                        goto out;
3332                r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3333                break;
3334        }
3335#endif
3336        case KVM_IRQFD: {
3337                struct kvm_irqfd data;
3338
3339                r = -EFAULT;
3340                if (copy_from_user(&data, argp, sizeof(data)))
3341                        goto out;
3342                r = kvm_irqfd(kvm, &data);
3343                break;
3344        }
3345        case KVM_IOEVENTFD: {
3346                struct kvm_ioeventfd data;
3347
3348                r = -EFAULT;
3349                if (copy_from_user(&data, argp, sizeof(data)))
3350                        goto out;
3351                r = kvm_ioeventfd(kvm, &data);
3352                break;
3353        }
3354#ifdef CONFIG_HAVE_KVM_MSI
3355        case KVM_SIGNAL_MSI: {
3356                struct kvm_msi msi;
3357
3358                r = -EFAULT;
3359                if (copy_from_user(&msi, argp, sizeof(msi)))
3360                        goto out;
3361                r = kvm_send_userspace_msi(kvm, &msi);
3362                break;
3363        }
3364#endif
3365#ifdef __KVM_HAVE_IRQ_LINE
3366        case KVM_IRQ_LINE_STATUS:
3367        case KVM_IRQ_LINE: {
3368                struct kvm_irq_level irq_event;
3369
3370                r = -EFAULT;
3371                if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3372                        goto out;
3373
3374                r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3375                                        ioctl == KVM_IRQ_LINE_STATUS);
3376                if (r)
3377                        goto out;
3378
3379                r = -EFAULT;
3380                if (ioctl == KVM_IRQ_LINE_STATUS) {
3381                        if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3382                                goto out;
3383                }
3384
3385                r = 0;
3386                break;
3387        }
3388#endif
3389#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3390        case KVM_SET_GSI_ROUTING: {
3391                struct kvm_irq_routing routing;
3392                struct kvm_irq_routing __user *urouting;
3393                struct kvm_irq_routing_entry *entries = NULL;
3394
3395                r = -EFAULT;
3396                if (copy_from_user(&routing, argp, sizeof(routing)))
3397                        goto out;
3398                r = -EINVAL;
3399                if (!kvm_arch_can_set_irq_routing(kvm))
3400                        goto out;
3401                if (routing.nr > KVM_MAX_IRQ_ROUTES)
3402                        goto out;
3403                if (routing.flags)
3404                        goto out;
3405                if (routing.nr) {
3406                        r = -ENOMEM;
3407                        entries = vmalloc(array_size(sizeof(*entries),
3408                                                     routing.nr));
3409                        if (!entries)
3410                                goto out;
3411                        r = -EFAULT;
3412                        urouting = argp;
3413                        if (copy_from_user(entries, urouting->entries,
3414                                           routing.nr * sizeof(*entries)))
3415                                goto out_free_irq_routing;
3416                }
3417                r = kvm_set_irq_routing(kvm, entries, routing.nr,
3418                                        routing.flags);
3419out_free_irq_routing:
3420                vfree(entries);
3421                break;
3422        }
3423#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3424        case KVM_CREATE_DEVICE: {
3425                struct kvm_create_device cd;
3426
3427                r = -EFAULT;
3428                if (copy_from_user(&cd, argp, sizeof(cd)))
3429                        goto out;
3430
3431                r = kvm_ioctl_create_device(kvm, &cd);
3432                if (r)
3433                        goto out;
3434
3435                r = -EFAULT;
3436                if (copy_to_user(argp, &cd, sizeof(cd)))
3437                        goto out;
3438
3439                r = 0;
3440                break;
3441        }
3442        case KVM_CHECK_EXTENSION:
3443                r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3444                break;
3445        default:
3446                r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3447        }
3448out:
3449        return r;
3450}
3451
3452#ifdef CONFIG_KVM_COMPAT
3453struct compat_kvm_dirty_log {
3454        __u32 slot;
3455        __u32 padding1;
3456        union {
3457                compat_uptr_t dirty_bitmap; /* one bit per page */
3458                __u64 padding2;
3459        };
3460};
3461
3462static long kvm_vm_compat_ioctl(struct file *filp,
3463                           unsigned int ioctl, unsigned long arg)
3464{
3465        struct kvm *kvm = filp->private_data;
3466        int r;
3467
3468        if (kvm->mm != current->mm)
3469                return -EIO;
3470        switch (ioctl) {
3471        case KVM_GET_DIRTY_LOG: {
3472                struct compat_kvm_dirty_log compat_log;
3473                struct kvm_dirty_log log;
3474
3475                if (copy_from_user(&compat_log, (void __user *)arg,
3476                                   sizeof(compat_log)))
3477                        return -EFAULT;
3478                log.slot         = compat_log.slot;
3479                log.padding1     = compat_log.padding1;
3480                log.padding2     = compat_log.padding2;
3481                log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3482
3483                r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3484                break;
3485        }
3486        default:
3487                r = kvm_vm_ioctl(filp, ioctl, arg);
3488        }
3489        return r;
3490}
3491#endif
3492
3493static struct file_operations kvm_vm_fops = {
3494        .release        = kvm_vm_release,
3495        .unlocked_ioctl = kvm_vm_ioctl,
3496        .llseek         = noop_llseek,
3497        KVM_COMPAT(kvm_vm_compat_ioctl),
3498};
3499
3500static int kvm_dev_ioctl_create_vm(unsigned long type)
3501{
3502        int r;
3503        struct kvm *kvm;
3504        struct file *file;
3505
3506        kvm = kvm_create_vm(type);
3507        if (IS_ERR(kvm))
3508                return PTR_ERR(kvm);
3509#ifdef CONFIG_KVM_MMIO
3510        r = kvm_coalesced_mmio_init(kvm);
3511        if (r < 0)
3512                goto put_kvm;
3513#endif
3514        r = get_unused_fd_flags(O_CLOEXEC);
3515        if (r < 0)
3516                goto put_kvm;
3517
3518        file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3519        if (IS_ERR(file)) {
3520                put_unused_fd(r);
3521                r = PTR_ERR(file);
3522                goto put_kvm;
3523        }
3524
3525        /*
3526         * Don't call kvm_put_kvm anymore at this point; file->f_op is
3527         * already set, with ->release() being kvm_vm_release().  In error
3528         * cases it will be called by the final fput(file) and will take
3529         * care of doing kvm_put_kvm(kvm).
3530         */
3531        if (kvm_create_vm_debugfs(kvm, r) < 0) {
3532                put_unused_fd(r);
3533                fput(file);
3534                return -ENOMEM;
3535        }
3536        kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3537
3538        fd_install(r, file);
3539        return r;
3540
3541put_kvm:
3542        kvm_put_kvm(kvm);
3543        return r;
3544}
3545
3546static long kvm_dev_ioctl(struct file *filp,
3547                          unsigned int ioctl, unsigned long arg)
3548{
3549        long r = -EINVAL;
3550
3551        switch (ioctl) {
3552        case KVM_GET_API_VERSION:
3553                if (arg)
3554                        goto out;
3555                r = KVM_API_VERSION;
3556                break;
3557        case KVM_CREATE_VM:
3558                r = kvm_dev_ioctl_create_vm(arg);
3559                break;
3560        case KVM_CHECK_EXTENSION:
3561                r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3562                break;
3563        case KVM_GET_VCPU_MMAP_SIZE:
3564                if (arg)
3565                        goto out;
3566                r = PAGE_SIZE;     /* struct kvm_run */
3567#ifdef CONFIG_X86
3568                r += PAGE_SIZE;    /* pio data page */
3569#endif
3570#ifdef CONFIG_KVM_MMIO
3571                r += PAGE_SIZE;    /* coalesced mmio ring page */
3572#endif
3573                break;
3574        case KVM_TRACE_ENABLE:
3575        case KVM_TRACE_PAUSE:
3576        case KVM_TRACE_DISABLE:
3577                r = -EOPNOTSUPP;
3578                break;
3579        default:
3580                return kvm_arch_dev_ioctl(filp, ioctl, arg);
3581        }
3582out:
3583        return r;
3584}
3585
3586static struct file_operations kvm_chardev_ops = {
3587        .unlocked_ioctl = kvm_dev_ioctl,
3588        .llseek         = noop_llseek,
3589        KVM_COMPAT(kvm_dev_ioctl),
3590};
3591
3592static struct miscdevice kvm_dev = {
3593        KVM_MINOR,
3594        "kvm",
3595        &kvm_chardev_ops,
3596};
3597
3598static void hardware_enable_nolock(void *junk)
3599{
3600        int cpu = raw_smp_processor_id();
3601        int r;
3602
3603        if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3604                return;
3605
3606        cpumask_set_cpu(cpu, cpus_hardware_enabled);
3607
3608        r = kvm_arch_hardware_enable();
3609
3610        if (r) {
3611                cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3612                atomic_inc(&hardware_enable_failed);
3613                pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3614        }
3615}
3616
3617static int kvm_starting_cpu(unsigned int cpu)
3618{
3619        raw_spin_lock(&kvm_count_lock);
3620        if (kvm_usage_count)
3621                hardware_enable_nolock(NULL);
3622        raw_spin_unlock(&kvm_count_lock);
3623        return 0;
3624}
3625
3626static void hardware_disable_nolock(void *junk)
3627{
3628        int cpu = raw_smp_processor_id();
3629
3630        if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3631                return;
3632        cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3633        kvm_arch_hardware_disable();
3634}
3635
3636static int kvm_dying_cpu(unsigned int cpu)
3637{
3638        raw_spin_lock(&kvm_count_lock);
3639        if (kvm_usage_count)
3640                hardware_disable_nolock(NULL);
3641        raw_spin_unlock(&kvm_count_lock);
3642        return 0;
3643}
3644
3645static void hardware_disable_all_nolock(void)
3646{
3647        BUG_ON(!kvm_usage_count);
3648
3649        kvm_usage_count--;
3650        if (!kvm_usage_count)
3651                on_each_cpu(hardware_disable_nolock, NULL, 1);
3652}
3653
3654static void hardware_disable_all(void)
3655{
3656        raw_spin_lock(&kvm_count_lock);
3657        hardware_disable_all_nolock();
3658        raw_spin_unlock(&kvm_count_lock);
3659}
3660
3661static int hardware_enable_all(void)
3662{
3663        int r = 0;
3664
3665        raw_spin_lock(&kvm_count_lock);
3666
3667        kvm_usage_count++;
3668        if (kvm_usage_count == 1) {
3669                atomic_set(&hardware_enable_failed, 0);
3670                on_each_cpu(hardware_enable_nolock, NULL, 1);
3671
3672                if (atomic_read(&hardware_enable_failed)) {
3673                        hardware_disable_all_nolock();
3674                        r = -EBUSY;
3675                }
3676        }
3677
3678        raw_spin_unlock(&kvm_count_lock);
3679
3680        return r;
3681}
3682
3683static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3684                      void *v)
3685{
3686        /*
3687         * Some (well, at least mine) BIOSes hang on reboot if
3688         * in vmx root mode.
3689         *
3690         * And Intel TXT required VMX off for all cpu when system shutdown.
3691         */
3692        pr_info("kvm: exiting hardware virtualization\n");
3693        kvm_rebooting = true;
3694        on_each_cpu(hardware_disable_nolock, NULL, 1);
3695        return NOTIFY_OK;
3696}
3697
3698static struct notifier_block kvm_reboot_notifier = {
3699        .notifier_call = kvm_reboot,
3700        .priority = 0,
3701};
3702
3703static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3704{
3705        int i;
3706
3707        for (i = 0; i < bus->dev_count; i++) {
3708                struct kvm_io_device *pos = bus->range[i].dev;
3709
3710                kvm_iodevice_destructor(pos);
3711        }
3712        kfree(bus);
3713}
3714
3715static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3716                                 const struct kvm_io_range *r2)
3717{
3718        gpa_t addr1 = r1->addr;
3719        gpa_t addr2 = r2->addr;
3720
3721        if (addr1 < addr2)
3722                return -1;
3723
3724        /* If r2->len == 0, match the exact address.  If r2->len != 0,
3725         * accept any overlapping write.  Any order is acceptable for
3726         * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3727         * we process all of them.
3728         */
3729        if (r2->len) {
3730                addr1 += r1->len;
3731                addr2 += r2->len;
3732        }
3733
3734        if (addr1 > addr2)
3735                return 1;
3736
3737        return 0;
3738}
3739
3740static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3741{
3742        return kvm_io_bus_cmp(p1, p2);
3743}
3744
3745static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3746                             gpa_t addr, int len)
3747{
3748        struct kvm_io_range *range, key;
3749        int off;
3750
3751        key = (struct kvm_io_range) {
3752                .addr = addr,
3753                .len = len,
3754        };
3755
3756        range = bsearch(&key, bus->range, bus->dev_count,
3757                        sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3758        if (range == NULL)
3759                return -ENOENT;
3760
3761        off = range - bus->range;
3762
3763        while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3764                off--;
3765
3766        return off;
3767}
3768
3769static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3770                              struct kvm_io_range *range, const void *val)
3771{
3772        int idx;
3773
3774        idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3775        if (idx < 0)
3776                return -EOPNOTSUPP;
3777
3778        while (idx < bus->dev_count &&
3779                kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3780                if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3781                                        range->len, val))
3782                        return idx;
3783                idx++;
3784        }
3785
3786        return -EOPNOTSUPP;
3787}
3788
3789/* kvm_io_bus_write - called under kvm->slots_lock */
3790int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3791                     int len, const void *val)
3792{
3793        struct kvm_io_bus *bus;
3794        struct kvm_io_range range;
3795        int r;
3796
3797        range = (struct kvm_io_range) {
3798                .addr = addr,
3799                .len = len,
3800        };
3801
3802        bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3803        if (!bus)
3804                return -ENOMEM;
3805        r = __kvm_io_bus_write(vcpu, bus, &range, val);
3806        return r < 0 ? r : 0;
3807}
3808EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3809
3810/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3811int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3812                            gpa_t addr, int len, const void *val, long cookie)
3813{
3814        struct kvm_io_bus *bus;
3815        struct kvm_io_range range;
3816
3817        range = (struct kvm_io_range) {
3818                .addr = addr,
3819                .len = len,
3820        };
3821
3822        bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3823        if (!bus)
3824                return -ENOMEM;
3825
3826        /* First try the device referenced by cookie. */
3827        if ((cookie >= 0) && (cookie < bus->dev_count) &&
3828            (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3829                if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3830                                        val))
3831                        return cookie;
3832
3833        /*
3834         * cookie contained garbage; fall back to search and return the
3835         * correct cookie value.
3836         */
3837        return __kvm_io_bus_write(vcpu, bus, &range, val);
3838}
3839
3840static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3841                             struct kvm_io_range *range, void *val)
3842{
3843        int idx;
3844
3845        idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3846        if (idx < 0)
3847                return -EOPNOTSUPP;
3848
3849        while (idx < bus->dev_count &&
3850                kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3851                if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3852                                       range->len, val))
3853                        return idx;
3854                idx++;
3855        }
3856
3857        return -EOPNOTSUPP;
3858}
3859
3860/* kvm_io_bus_read - called under kvm->slots_lock */
3861int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3862                    int len, void *val)
3863{
3864        struct kvm_io_bus *bus;
3865        struct kvm_io_range range;
3866        int r;
3867
3868        range = (struct kvm_io_range) {
3869                .addr = addr,
3870                .len = len,
3871        };
3872
3873        bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3874        if (!bus)
3875                return -ENOMEM;
3876        r = __kvm_io_bus_read(vcpu, bus, &range, val);
3877        return r < 0 ? r : 0;
3878}
3879
3880/* Caller must hold slots_lock. */
3881int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3882                            int len, struct kvm_io_device *dev)
3883{
3884        int i;
3885        struct kvm_io_bus *new_bus, *bus;
3886        struct kvm_io_range range;
3887
3888        bus = kvm_get_bus(kvm, bus_idx);
3889        if (!bus)
3890                return -ENOMEM;
3891
3892        /* exclude ioeventfd which is limited by maximum fd */
3893        if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3894                return -ENOSPC;
3895
3896        new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
3897                          GFP_KERNEL_ACCOUNT);
3898        if (!new_bus)
3899                return -ENOMEM;
3900
3901        range = (struct kvm_io_range) {
3902                .addr = addr,
3903                .len = len,
3904                .dev = dev,
3905        };
3906
3907        for (i = 0; i < bus->dev_count; i++)
3908                if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
3909                        break;
3910
3911        memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3912        new_bus->dev_count++;
3913        new_bus->range[i] = range;
3914        memcpy(new_bus->range + i + 1, bus->range + i,
3915                (bus->dev_count - i) * sizeof(struct kvm_io_range));
3916        rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3917        synchronize_srcu_expedited(&kvm->srcu);
3918        kfree(bus);
3919
3920        return 0;
3921}
3922
3923/* Caller must hold slots_lock. */
3924void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3925                               struct kvm_io_device *dev)
3926{
3927        int i;
3928        struct kvm_io_bus *new_bus, *bus;
3929
3930        bus = kvm_get_bus(kvm, bus_idx);
3931        if (!bus)
3932                return;
3933
3934        for (i = 0; i < bus->dev_count; i++)
3935                if (bus->range[i].dev == dev) {
3936                        break;
3937                }
3938
3939        if (i == bus->dev_count)
3940                return;
3941
3942        new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
3943                          GFP_KERNEL_ACCOUNT);
3944        if (!new_bus)  {
3945                pr_err("kvm: failed to shrink bus, removing it completely\n");
3946                goto broken;
3947        }
3948
3949        memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3950        new_bus->dev_count--;
3951        memcpy(new_bus->range + i, bus->range + i + 1,
3952               (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3953
3954broken:
3955        rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3956        synchronize_srcu_expedited(&kvm->srcu);
3957        kfree(bus);
3958        return;
3959}
3960
3961struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3962                                         gpa_t addr)
3963{
3964        struct kvm_io_bus *bus;
3965        int dev_idx, srcu_idx;
3966        struct kvm_io_device *iodev = NULL;
3967
3968        srcu_idx = srcu_read_lock(&kvm->srcu);
3969
3970        bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3971        if (!bus)
3972                goto out_unlock;
3973
3974        dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3975        if (dev_idx < 0)
3976                goto out_unlock;
3977
3978        iodev = bus->range[dev_idx].dev;
3979
3980out_unlock:
3981        srcu_read_unlock(&kvm->srcu, srcu_idx);
3982
3983        return iodev;
3984}
3985EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3986
3987static int kvm_debugfs_open(struct inode *inode, struct file *file,
3988                           int (*get)(void *, u64 *), int (*set)(void *, u64),
3989                           const char *fmt)
3990{
3991        struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3992                                          inode->i_private;
3993
3994        /* The debugfs files are a reference to the kvm struct which
3995         * is still valid when kvm_destroy_vm is called.
3996         * To avoid the race between open and the removal of the debugfs
3997         * directory we test against the users count.
3998         */
3999        if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
4000                return -ENOENT;
4001
4002        if (simple_attr_open(inode, file, get,
4003                             stat_data->mode & S_IWUGO ? set : NULL,
4004                             fmt)) {
4005                kvm_put_kvm(stat_data->kvm);
4006                return -ENOMEM;
4007        }
4008
4009        return 0;
4010}
4011
4012static int kvm_debugfs_release(struct inode *inode, struct file *file)
4013{
4014        struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
4015                                          inode->i_private;
4016
4017        simple_attr_release(inode, file);
4018        kvm_put_kvm(stat_data->kvm);
4019
4020        return 0;
4021}
4022
4023static int vm_stat_get_per_vm(void *data, u64 *val)
4024{
4025        struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4026
4027        *val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
4028
4029        return 0;
4030}
4031
4032static int vm_stat_clear_per_vm(void *data, u64 val)
4033{
4034        struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4035
4036        if (val)
4037                return -EINVAL;
4038
4039        *(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
4040
4041        return 0;
4042}
4043
4044static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
4045{
4046        __simple_attr_check_format("%llu\n", 0ull);
4047        return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
4048                                vm_stat_clear_per_vm, "%llu\n");
4049}
4050
4051static const struct file_operations vm_stat_get_per_vm_fops = {
4052        .owner   = THIS_MODULE,
4053        .open    = vm_stat_get_per_vm_open,
4054        .release = kvm_debugfs_release,
4055        .read    = simple_attr_read,
4056        .write   = simple_attr_write,
4057        .llseek  = no_llseek,
4058};
4059
4060static int vcpu_stat_get_per_vm(void *data, u64 *val)
4061{
4062        int i;
4063        struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4064        struct kvm_vcpu *vcpu;
4065
4066        *val = 0;
4067
4068        kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4069                *val += *(u64 *)((void *)vcpu + stat_data->offset);
4070
4071        return 0;
4072}
4073
4074static int vcpu_stat_clear_per_vm(void *data, u64 val)
4075{
4076        int i;
4077        struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
4078        struct kvm_vcpu *vcpu;
4079
4080        if (val)
4081                return -EINVAL;
4082
4083        kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
4084                *(u64 *)((void *)vcpu + stat_data->offset) = 0;
4085
4086        return 0;
4087}
4088
4089static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
4090{
4091        __simple_attr_check_format("%llu\n", 0ull);
4092        return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
4093                                 vcpu_stat_clear_per_vm, "%llu\n");
4094}
4095
4096static const struct file_operations vcpu_stat_get_per_vm_fops = {
4097        .owner   = THIS_MODULE,
4098        .open    = vcpu_stat_get_per_vm_open,
4099        .release = kvm_debugfs_release,
4100        .read    = simple_attr_read,
4101        .write   = simple_attr_write,
4102        .llseek  = no_llseek,
4103};
4104
4105static const struct file_operations *stat_fops_per_vm[] = {
4106        [KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
4107        [KVM_STAT_VM]   = &vm_stat_get_per_vm_fops,
4108};
4109
4110static int vm_stat_get(void *_offset, u64 *val)
4111{
4112        unsigned offset = (long)_offset;
4113        struct kvm *kvm;
4114        struct kvm_stat_data stat_tmp = {.offset = offset};
4115        u64 tmp_val;
4116
4117        *val = 0;
4118        mutex_lock(&kvm_lock);
4119        list_for_each_entry(kvm, &vm_list, vm_list) {
4120                stat_tmp.kvm = kvm;
4121                vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4122                *val += tmp_val;
4123        }
4124        mutex_unlock(&kvm_lock);
4125        return 0;
4126}
4127
4128static int vm_stat_clear(void *_offset, u64 val)
4129{
4130        unsigned offset = (long)_offset;
4131        struct kvm *kvm;
4132        struct kvm_stat_data stat_tmp = {.offset = offset};
4133
4134        if (val)
4135                return -EINVAL;
4136
4137        mutex_lock(&kvm_lock);
4138        list_for_each_entry(kvm, &vm_list, vm_list) {
4139                stat_tmp.kvm = kvm;
4140                vm_stat_clear_per_vm((void *)&stat_tmp, 0);
4141        }
4142        mutex_unlock(&kvm_lock);
4143
4144        return 0;
4145}
4146
4147DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
4148
4149static int vcpu_stat_get(void *_offset, u64 *val)
4150{
4151        unsigned offset = (long)_offset;
4152        struct kvm *kvm;
4153        struct kvm_stat_data stat_tmp = {.offset = offset};
4154        u64 tmp_val;
4155
4156        *val = 0;
4157        mutex_lock(&kvm_lock);
4158        list_for_each_entry(kvm, &vm_list, vm_list) {
4159                stat_tmp.kvm = kvm;
4160                vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
4161                *val += tmp_val;
4162        }
4163        mutex_unlock(&kvm_lock);
4164        return 0;
4165}
4166
4167static int vcpu_stat_clear(void *_offset, u64 val)
4168{
4169        unsigned offset = (long)_offset;
4170        struct kvm *kvm;
4171        struct kvm_stat_data stat_tmp = {.offset = offset};
4172
4173        if (val)
4174                return -EINVAL;
4175
4176        mutex_lock(&kvm_lock);
4177        list_for_each_entry(kvm, &vm_list, vm_list) {
4178                stat_tmp.kvm = kvm;
4179                vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
4180        }
4181        mutex_unlock(&kvm_lock);
4182
4183        return 0;
4184}
4185
4186DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
4187                        "%llu\n");
4188
4189static const struct file_operations *stat_fops[] = {
4190        [KVM_STAT_VCPU] = &vcpu_stat_fops,
4191        [KVM_STAT_VM]   = &vm_stat_fops,
4192};
4193
4194static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
4195{
4196        struct kobj_uevent_env *env;
4197        unsigned long long created, active;
4198
4199        if (!kvm_dev.this_device || !kvm)
4200                return;
4201
4202        mutex_lock(&kvm_lock);
4203        if (type == KVM_EVENT_CREATE_VM) {
4204                kvm_createvm_count++;
4205                kvm_active_vms++;
4206        } else if (type == KVM_EVENT_DESTROY_VM) {
4207                kvm_active_vms--;
4208        }
4209        created = kvm_createvm_count;
4210        active = kvm_active_vms;
4211        mutex_unlock(&kvm_lock);
4212
4213        env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
4214        if (!env)
4215                return;
4216
4217        add_uevent_var(env, "CREATED=%llu", created);
4218        add_uevent_var(env, "COUNT=%llu", active);
4219
4220        if (type == KVM_EVENT_CREATE_VM) {
4221                add_uevent_var(env, "EVENT=create");
4222                kvm->userspace_pid = task_pid_nr(current);
4223        } else if (type == KVM_EVENT_DESTROY_VM) {
4224                add_uevent_var(env, "EVENT=destroy");
4225        }
4226        add_uevent_var(env, "PID=%d", kvm->userspace_pid);
4227
4228        if (!IS_ERR_OR_NULL(kvm->debugfs_dentry)) {
4229                char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
4230
4231                if (p) {
4232                        tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
4233                        if (!IS_ERR(tmp))
4234                                add_uevent_var(env, "STATS_PATH=%s", tmp);
4235                        kfree(p);
4236                }
4237        }
4238        /* no need for checks, since we are adding at most only 5 keys */
4239        env->envp[env->envp_idx++] = NULL;
4240        kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
4241        kfree(env);
4242}
4243
4244static void kvm_init_debug(void)
4245{
4246        struct kvm_stats_debugfs_item *p;
4247
4248        kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
4249
4250        kvm_debugfs_num_entries = 0;
4251        for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
4252                int mode = p->mode ? p->mode : 0644;
4253                debugfs_create_file(p->name, mode, kvm_debugfs_dir,
4254                                    (void *)(long)p->offset,
4255                                    stat_fops[p->kind]);
4256        }
4257}
4258
4259static int kvm_suspend(void)
4260{
4261        if (kvm_usage_count)
4262                hardware_disable_nolock(NULL);
4263        return 0;
4264}
4265
4266static void kvm_resume(void)
4267{
4268        if (kvm_usage_count) {
4269#ifdef CONFIG_LOCKDEP
4270                WARN_ON(lockdep_is_held(&kvm_count_lock));
4271#endif
4272                hardware_enable_nolock(NULL);
4273        }
4274}
4275
4276static struct syscore_ops kvm_syscore_ops = {
4277        .suspend = kvm_suspend,
4278        .resume = kvm_resume,
4279};
4280
4281static inline
4282struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
4283{
4284        return container_of(pn, struct kvm_vcpu, preempt_notifier);
4285}
4286
4287static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
4288{
4289        struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4290
4291        WRITE_ONCE(vcpu->preempted, false);
4292        WRITE_ONCE(vcpu->ready, false);
4293
4294        kvm_arch_sched_in(vcpu, cpu);
4295
4296        kvm_arch_vcpu_load(vcpu, cpu);
4297}
4298
4299static void kvm_sched_out(struct preempt_notifier *pn,
4300                          struct task_struct *next)
4301{
4302        struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4303
4304        if (current->state == TASK_RUNNING) {
4305                WRITE_ONCE(vcpu->preempted, true);
4306                WRITE_ONCE(vcpu->ready, true);
4307        }
4308        kvm_arch_vcpu_put(vcpu);
4309}
4310
4311static void check_processor_compat(void *rtn)
4312{
4313        *(int *)rtn = kvm_arch_check_processor_compat();
4314}
4315
4316int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4317                  struct module *module)
4318{
4319        int r;
4320        int cpu;
4321
4322        r = kvm_arch_init(opaque);
4323        if (r)
4324                goto out_fail;
4325
4326        /*
4327         * kvm_arch_init makes sure there's at most one caller
4328         * for architectures that support multiple implementations,
4329         * like intel and amd on x86.
4330         * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4331         * conflicts in case kvm is already setup for another implementation.
4332         */
4333        r = kvm_irqfd_init();
4334        if (r)
4335                goto out_irqfd;
4336
4337        if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4338                r = -ENOMEM;
4339                goto out_free_0;
4340        }
4341
4342        r = kvm_arch_hardware_setup();
4343        if (r < 0)
4344                goto out_free_0a;
4345
4346        for_each_online_cpu(cpu) {
4347                smp_call_function_single(cpu, check_processor_compat, &r, 1);
4348                if (r < 0)
4349                        goto out_free_1;
4350        }
4351
4352        r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4353                                      kvm_starting_cpu, kvm_dying_cpu);
4354        if (r)
4355                goto out_free_2;
4356        register_reboot_notifier(&kvm_reboot_notifier);
4357
4358        /* A kmem cache lets us meet the alignment requirements of fx_save. */
4359        if (!vcpu_align)
4360                vcpu_align = __alignof__(struct kvm_vcpu);
4361        kvm_vcpu_cache =
4362                kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
4363                                           SLAB_ACCOUNT,
4364                                           offsetof(struct kvm_vcpu, arch),
4365                                           sizeof_field(struct kvm_vcpu, arch),
4366                                           NULL);
4367        if (!kvm_vcpu_cache) {
4368                r = -ENOMEM;
4369                goto out_free_3;
4370        }
4371
4372        r = kvm_async_pf_init();
4373        if (r)
4374                goto out_free;
4375
4376        kvm_chardev_ops.owner = module;
4377        kvm_vm_fops.owner = module;
4378        kvm_vcpu_fops.owner = module;
4379
4380        r = misc_register(&kvm_dev);
4381        if (r) {
4382                pr_err("kvm: misc device register failed\n");
4383                goto out_unreg;
4384        }
4385
4386        register_syscore_ops(&kvm_syscore_ops);
4387
4388        kvm_preempt_ops.sched_in = kvm_sched_in;
4389        kvm_preempt_ops.sched_out = kvm_sched_out;
4390
4391        kvm_init_debug();
4392
4393        r = kvm_vfio_ops_init();
4394        WARN_ON(r);
4395
4396        return 0;
4397
4398out_unreg:
4399        kvm_async_pf_deinit();
4400out_free:
4401        kmem_cache_destroy(kvm_vcpu_cache);
4402out_free_3:
4403        unregister_reboot_notifier(&kvm_reboot_notifier);
4404        cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4405out_free_2:
4406out_free_1:
4407        kvm_arch_hardware_unsetup();
4408out_free_0a:
4409        free_cpumask_var(cpus_hardware_enabled);
4410out_free_0:
4411        kvm_irqfd_exit();
4412out_irqfd:
4413        kvm_arch_exit();
4414out_fail:
4415        return r;
4416}
4417EXPORT_SYMBOL_GPL(kvm_init);
4418
4419void kvm_exit(void)
4420{
4421        debugfs_remove_recursive(kvm_debugfs_dir);
4422        misc_deregister(&kvm_dev);
4423        kmem_cache_destroy(kvm_vcpu_cache);
4424        kvm_async_pf_deinit();
4425        unregister_syscore_ops(&kvm_syscore_ops);
4426        unregister_reboot_notifier(&kvm_reboot_notifier);
4427        cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4428        on_each_cpu(hardware_disable_nolock, NULL, 1);
4429        kvm_arch_hardware_unsetup();
4430        kvm_arch_exit();
4431        kvm_irqfd_exit();
4432        free_cpumask_var(cpus_hardware_enabled);
4433        kvm_vfio_ops_exit();
4434}
4435EXPORT_SYMBOL_GPL(kvm_exit);
4436
4437struct kvm_vm_worker_thread_context {
4438        struct kvm *kvm;
4439        struct task_struct *parent;
4440        struct completion init_done;
4441        kvm_vm_thread_fn_t thread_fn;
4442        uintptr_t data;
4443        int err;
4444};
4445
4446static int kvm_vm_worker_thread(void *context)
4447{
4448        /*
4449         * The init_context is allocated on the stack of the parent thread, so
4450         * we have to locally copy anything that is needed beyond initialization
4451         */
4452        struct kvm_vm_worker_thread_context *init_context = context;
4453        struct kvm *kvm = init_context->kvm;
4454        kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
4455        uintptr_t data = init_context->data;
4456        int err;
4457
4458        err = kthread_park(current);
4459        /* kthread_park(current) is never supposed to return an error */
4460        WARN_ON(err != 0);
4461        if (err)
4462                goto init_complete;
4463
4464        err = cgroup_attach_task_all(init_context->parent, current);
4465        if (err) {
4466                kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
4467                        __func__, err);
4468                goto init_complete;
4469        }
4470
4471        set_user_nice(current, task_nice(init_context->parent));
4472
4473init_complete:
4474        init_context->err = err;
4475        complete(&init_context->init_done);
4476        init_context = NULL;
4477
4478        if (err)
4479                return err;
4480
4481        /* Wait to be woken up by the spawner before proceeding. */
4482        kthread_parkme();
4483
4484        if (!kthread_should_stop())
4485                err = thread_fn(kvm, data);
4486
4487        return err;
4488}
4489
4490int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
4491                                uintptr_t data, const char *name,
4492                                struct task_struct **thread_ptr)
4493{
4494        struct kvm_vm_worker_thread_context init_context = {};
4495        struct task_struct *thread;
4496
4497        *thread_ptr = NULL;
4498        init_context.kvm = kvm;
4499        init_context.parent = current;
4500        init_context.thread_fn = thread_fn;
4501        init_context.data = data;
4502        init_completion(&init_context.init_done);
4503
4504        thread = kthread_run(kvm_vm_worker_thread, &init_context,
4505                             "%s-%d", name, task_pid_nr(current));
4506        if (IS_ERR(thread))
4507                return PTR_ERR(thread);
4508
4509        /* kthread_run is never supposed to return NULL */
4510        WARN_ON(thread == NULL);
4511
4512        wait_for_completion(&init_context.init_done);
4513
4514        if (!init_context.err)
4515                *thread_ptr = thread;
4516
4517        return init_context.err;
4518}
4519