Linux KVM虚拟机的创建

staticintkvm_dev_ioctl_create_vm(unsigned long type){    char fdname[ITOA_MAX_LEN + 1];    int r, fd;    struct kvm *kvm;    struct file *file;    /* ========== 步骤1: 分配文件描述符 ========== */    fd = get_unused_fd_flags(O_CLOEXEC);    /* ========== 步骤2: 生成文件描述符名称 ========== */    snprintf(fdname, sizeof(fdname), "%d", fd);    /* ========== 步骤3: 创建VM核心对象 ========== */    kvm = kvm_create_vm(type, fdname);    /* ========== 步骤4: 创建匿名inode文件 ========== */    file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);    /* ========== 步骤5: 发送uevent通知 ========== */    /*     * Don't call kvm_put_kvm anymore at this point; file->f_op is     * already set, with ->release() being kvm_vm_release().  In error     * cases it will be called by the final fput(file) and will take     * care of doing kvm_put_kvm(kvm).     */    kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);    /* ========== 步骤6: 安装文件描述符 ========== */    fd_install(fd, file);    return fd;    /* ========== 错误处理 ========== */}

static struct kvm *kvm_create_vm(unsigned long type, const char *fdname){    struct kvm *kvm = kvm_arch_alloc_vm();    struct kvm_memslots *slots;    int r, i, j;    /* ===== 第一部分: 基础初始化 ===== */    KVM_MMU_LOCK_INIT(kvm);    mmgrab(current->mm);    kvm->mm = current->mm;    kvm_eventfd_init(kvm);    mutex_init(&kvm->lock);    mutex_init(&kvm->irq_lock);    mutex_init(&kvm->slots_lock);    mutex_init(&kvm->slots_arch_lock);    spin_lock_init(&kvm->mn_invalidate_lock);    rcuwait_init(&kvm->mn_memslots_update_rcuwait);    xa_init(&kvm->vcpu_array);#ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES    xa_init(&kvm->mem_attr_array);#endif    INIT_LIST_HEAD(&kvm->gpc_list);    spin_lock_init(&kvm->gpc_lock);    INIT_LIST_HEAD(&kvm->devices);    kvm->max_vcpus = KVM_MAX_VCPUS;    BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);    /*     * Force subsequent debugfs file creations to fail if the VM directory     * is not created (by kvm_create_vm_debugfs()).     */    kvm->debugfs_dentry = ERR_PTR(-ENOENT);    snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d",         task_pid_nr(current));    /* ===== 第二部分: SRCU初始化 ===== */    r = -ENOMEM;    if (init_srcu_struct(&kvm->srcu))        goto out_err_no_srcu;    if (init_srcu_struct(&kvm->irq_srcu))        goto out_err_no_irq_srcu;    /* ===== 第三部分: IRQ路由初始化 ===== */    r = kvm_init_irq_routing(kvm);    /* ===== 第四部分: 引用计数初始化 ===== */    refcount_set(&kvm->users_count, 1);    /* ===== 第五部分: 内存槽初始化 ===== */    for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {        for (j = 0; j < 2; j++) {            slots = &kvm->__memslots[i][j];            atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);            slots->hva_tree = RB_ROOT_CACHED;            slots->gfn_tree = RB_ROOT;            hash_init(slots->id_hash);            slots->node_idx = j;            /* Generations must be different for each address space. */            slots->generation = i;        }        rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);    }    /* ===== 第六部分: I/O总线初始化 ===== */    for (i = 0; i < KVM_NR_BUSES; i++) {        rcu_assign_pointer(kvm->buses[i],            kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));    }    /* ===== 第七部分: 架构特定初始化 ===== */    r = kvm_arch_init_vm(kvm, type);    /* ===== 第八部分: 启用虚拟化 ===== */    r = kvm_enable_virtualization();    /* ===== 第九部分: IRQ确认通知器 ===== */#ifdef CONFIG_HAVE_KVM_IRQCHIP    INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);#endif    /* ===== 第十部分: MMU通知器初始化 ===== */    r = kvm_init_mmu_notifier(kvm);    /* ===== 第十一部分: 合并MMIO初始化 ===== */    r = kvm_coalesced_mmio_init(kvm);    /* ===== 第十二部分: 创建debugfs ===== */    r = kvm_create_vm_debugfs(kvm, fdname);    /* ===== 第十三部分: 添加到全局VM列表 ===== */    mutex_lock(&kvm_lock);    list_add(&kvm->vm_list, &vm_list);    mutex_unlock(&kvm_lock);    /* ===== 第十四部分: 通知器注册 ===== */    preempt_notifier_inc();    kvm_init_pm_notifier(kvm);    return kvm;    /* ===== 错误处理路径 ===== */}/* * Allocate empty IRQ routing by default so that additional setup isn't needed * when userspace-driven IRQ routing is activated, and so that kvm->irq_routing * is guaranteed to be non-NULL. */intkvm_init_irq_routing(struct kvm *kvm){    struct kvm_irq_routing_table *new;    int chip_size;    new = kzalloc(struct_size(new, map, 1), GFP_KERNEL_ACCOUNT);    new->nr_rt_entries = 1;    chip_size = sizeof(int) * KVM_NR_IRQCHIPS * KVM_IRQCHIP_NUM_PINS;    memset(new->chip, -1, chip_size);    RCU_INIT_POINTER(kvm->irq_routing, new);    return 0;}intkvm_coalesced_mmio_init(struct kvm *kvm){    struct page *page;    page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);    kvm->coalesced_mmio_ring = page_address(page);    /*     * We're using this spinlock to sync access to the coalesced ring.     * The list doesn't need its own lock since device registration and     * unregistration should only happen when kvm->slots_lock is held.     */    spin_lock_init(&kvm->ring_lock);    INIT_LIST_HEAD(&kvm->coalesced_zones);    return 0;}

intkvm_arch_init_vm(struct kvm *kvm, unsigned long type){    int ret;    /* ===== 步骤1: 初始化配置锁 ===== */    mutex_init(&kvm->arch.config_lock);#ifdef CONFIG_LOCKDEP    /* Clue in lockdep that the config_lock must be taken inside kvm->lock */    mutex_lock(&kvm->lock);    mutex_lock(&kvm->arch.config_lock);    mutex_unlock(&kvm->arch.config_lock);    mutex_unlock(&kvm->lock);#endif    /* ===== 步骤2: 初始化嵌套虚拟化 ===== */    kvm_init_nested(kvm);    /* ===== 步骤3: 共享KVM结构到Hyp ===== */    ret = kvm_share_hyp(kvm, kvm + 1);    /* ===== 步骤4: 分配支持的CPU掩码 ===== */    if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL_ACCOUNT)) {        ret = -ENOMEM;        goto err_unshare_kvm;    }    cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);    /* ===== 步骤5: 初始化Stage-2 MMU ===== */    ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type);    /* ===== 步骤6: Protected KVM初始化 ===== */    if (is_protected_kvm_enabled()) {        /*         * If any failures occur after this is successful, make sure to         * call __pkvm_unreserve_vm to unreserve the VM in hyp.         */        ret = pkvm_init_host_vm(kvm);    }    /* ===== 步骤7: VGIC早期初始化 ===== */    kvm_vgic_early_init(kvm);    /* ===== 步骤8: Timer初始化 ===== */    kvm_timer_init_vm(kvm);    /* ===== 步骤9: 设置最大vCPU数 ===== */    /* The maximum number of VCPUs is limited by the host's GIC model */    kvm->max_vcpus = kvm_arm_default_max_vcpus();    /* ===== 步骤10: 初始化Hypercalls ===== */    kvm_arm_init_hypercalls(kvm);    /* ===== 步骤11: 清空vCPU特性位图 ===== */    bitmap_zero(kvm->arch.vcpu_features, KVM_VCPU_MAX_FEATURES);    return 0;    /* ===== 错误处理 ===== */}/** * kvm_vgic_early_init() - Initialize static VGIC VCPU data structures * @kvm: The VM whose VGIC districutor should be initialized * * Only do initialization of static structures that don't require any * allocation or sizing information from userspace.  vgic_init() called * kvm_vgic_dist_init() which takes care of the rest. */voidkvm_vgic_early_init(struct kvm *kvm){    struct vgic_dist *dist = &kvm->arch.vgic;    xa_init_flags(&dist->lpi_xa, XA_FLAGS_LOCK_IRQ);}voidkvm_timer_init_vm(struct kvm *kvm){    for (int i = 0; i < NR_KVM_TIMERS; i++)        kvm->arch.timer_data.ppi[i] = default_ppi[i];}voidkvm_arm_init_hypercalls(struct kvm *kvm){    struct kvm_smccc_features *smccc_feat = &kvm->arch.smccc_feat;    smccc_feat->std_bmap = KVM_ARM_SMCCC_STD_FEATURES;    smccc_feat->std_hyp_bmap = KVM_ARM_SMCCC_STD_HYP_FEATURES;    smccc_feat->vendor_hyp_bmap = KVM_ARM_SMCCC_VENDOR_HYP_FEATURES;    mt_init(&kvm->arch.smccc_filter);}

int kvm_init_stage2_mmu(struct kvm *kvm, struct kvm_s2_mmu *mmu, unsigned long type){    int cpu, err;    struct kvm_pgtable *pgt;    /* ========== 步骤1: 检查是否已初始化 ========== */    /*     * If we already have our page tables in place, and that the     * MMU context is the canonical one, we have a bug somewhere,     * as this is only supposed to ever happen once per VM.     *     * Otherwise, we're building nested page tables, and that's     * probably because userspace called KVM_ARM_VCPU_INIT more     * than once on the same vcpu. Since that's actually legal,     * don't kick a fuss and leave gracefully.     */    if (mmu->pgt != NULL) {        if (kvm_is_nested_s2_mmu(kvm, mmu))            return 0;        kvm_err("kvm_arch already initialized?\n");        return -EINVAL;    }    /* ========== 步骤2: 初始化IPA范围 ========== */    err = kvm_init_ipa_range(mmu, type);    /* ========== 步骤3: 分配页表结构 ========== */    pgt = kzalloc(sizeof(*pgt), GFP_KERNEL_ACCOUNT);    /* ========== 步骤4: 初始化页表 ========== */    mmu->arch = &kvm->arch;    err = KVM_PGT_FN(kvm_pgtable_stage2_init)(pgt, mmu, &kvm_s2_mm_ops);    /* ========== 步骤5: 关联页表到MMU ========== */    mmu->pgt = pgt;    if (is_protected_kvm_enabled())        return 0;    /* ========== 步骤6: 分配Per-CPU追踪 ========== */    mmu->last_vcpu_ran = alloc_percpu(typeof(*mmu->last_vcpu_ran));    /* ========== 步骤7: 初始化Per-CPU数据 ========== */    for_each_possible_cpu(cpu)        *per_cpu_ptr(mmu->last_vcpu_ran, cpu) = -1;    /* ========== 步骤8: 配置页分割 ========== */    /* The eager page splitting is disabled by default */    mmu->split_page_chunk_size = KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT;    mmu->split_page_cache.gfp_zero = __GFP_ZERO;    /* ========== 步骤9: 记录页表物理地址 ========== */    mmu->pgd_phys = __pa(pgt->pgd);    /* ========== 步骤10: 嵌套虚拟化初始化 ========== */    if (kvm_is_nested_s2_mmu(kvm, mmu))        kvm_init_nested_s2_mmu(mmu);    return 0;    /* ========== 错误处理 ========== */}

static int kvm_init_ipa_range(struct kvm_s2_mmu *mmu, unsigned long type){    u32 kvm_ipa_limit = get_kvm_ipa_limit();    u64 mmfr0, mmfr1;    u32 phys_shift;    /* ===== 步骤1: 验证type标志 ===== */    if (type & ~KVM_VM_TYPE_ARM_IPA_SIZE_MASK)        return -EINVAL;    /* ===== 步骤2: 确定物理地址位数 ===== */    phys_shift = KVM_VM_TYPE_ARM_IPA_SIZE(type);    if (is_protected_kvm_enabled()) {        // Protected KVM: 使用系统限制        phys_shift = kvm_ipa_limit;    } else if (phys_shift) {        // 用户指定IPA大小: 验证范围        if (phys_shift > kvm_ipa_limit ||            phys_shift < ARM64_MIN_PARANGE_BITS)            return -EINVAL;    } else {        // 未指定: 使用默认值        phys_shift = KVM_PHYS_SHIFT;        if (phys_shift > kvm_ipa_limit) {            pr_warn_once("%s using unsupported default IPA limit, upgrade your VMM\n",                     current->comm);            return -EINVAL;        }    }    /* ===== 步骤3: 读取CPU特性寄存器 ===== */    mmfr0 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);    mmfr1 = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);    /* ===== 步骤4: 计算VTCR_EL2值 ===== */    mmu->vtcr = kvm_get_vtcr(mmfr0, mmfr1, phys_shift);    return 0;}u64 kvm_get_vtcr(u64 mmfr0, u64 mmfr1, u32 phys_shift){    u64 vtcr = VTCR_EL2_FLAGS;    u8 lvls;    // 1. 设置物理地址大小    vtcr |= kvm_get_parange(mmfr0) << VTCR_EL2_PS_SHIFT;    // 2. 设置T0SZ (地址大小)    vtcr |= VTCR_EL2_T0SZ(phys_shift);    // 3. 设置页表级数    lvls = stage2_pgtable_levels(phys_shift);    vtcr |= VTCR_EL2_LVLS_TO_SL0(lvls);    // 4. 设置页粒度    if (PAGE_SIZE == SZ_64K)        vtcr |= VTCR_EL2_TG0_64K;    else if (PAGE_SIZE == SZ_16K)        vtcr |= VTCR_EL2_TG0_16K;    else        vtcr |= VTCR_EL2_TG0_4K;    // 5. 设置shareability和cacheability    vtcr |= VTCR_EL2_SH0_INNER | VTCR_EL2_ORGN0_WBWA |             VTCR_EL2_IRGN0_WBWA;    return vtcr;}

int __kvm_pgtable_stage2_init(struct kvm_pgtable *pgt, struct kvm_s2_mmu *mmu,                  struct kvm_pgtable_mm_ops *mm_ops,                  enum kvm_pgtable_stage2_flags flags,                  kvm_pgtable_force_pte_cb_t force_pte_cb){    size_t pgd_sz;    u64 vtcr = mmu->vtcr;    u32 ia_bits = VTCR_EL2_IPA(vtcr);    u32 sl0 = FIELD_GET(VTCR_EL2_SL0_MASK, vtcr);    s8 start_level = VTCR_EL2_TGRAN_SL0_BASE - sl0;    /* ===== 步骤1: 计算PGD大小 ===== */    pgd_sz = kvm_pgd_pages(ia_bits, start_level) * PAGE_SIZE;    /* ===== 步骤2: 分配PGD ===== */    pgt->pgd = (kvm_pteref_t)mm_ops->zalloc_pages_exact(pgd_sz);    /* ===== 步骤3: 初始化页表结构 ===== */    pgt->ia_bits        = ia_bits;    pgt->start_level    = start_level;    pgt->mm_ops     = mm_ops;    pgt->mmu        = mmu;    pgt->flags      = flags;    pgt->force_pte_cb   = force_pte_cb;    /* ===== 步骤4: 确保PGD对硬件可见 ===== */    /* Ensure zeroed PGD pages are visible to the hardware walker */    dsb(ishst);    return 0;}staticinline u32 kvm_pgd_pages(u32 ia_bits, s8 start_level){    // 根据IPA位数和起始级别计算需要多少页    return stage2_pgtable_levels(ia_bits) - start_level;}staticinline u32 stage2_pgtable_levels(u32 ipa_bits){    u32 bits_resolved = PAGE_SHIFT;    // 每级页表解析9位（对于4KB页）    if (ipa_bits > bits_resolved)        return DIV_ROUND_UP(ipa_bits - bits_resolved,                             PAGE_SHIFT - 3);    return 1;}