xref: /DragonOS/kernel/src/arch/x86_64/kvm/vmx/mmu.rs (revision cf7f801e1d50ee5b04cb728e4251a57f4183bfbc)

use crate::{
    arch::kvm::vmx::ept::EptMapper,
    libs::mutex::Mutex,
    mm::{page::EntryFlags, syscall::ProtFlags},
    virt::kvm::host_mem::{__gfn_to_pfn, kvm_vcpu_gfn_to_memslot, PAGE_MASK, PAGE_SHIFT},
};
use bitfield_struct::bitfield;
use log::debug;
use system_error::SystemError;

use super::{
    ept::check_ept_features,
    vcpu::VmxVcpu,
    vmcs::VmcsFields,
    vmx_asm_wrapper::{vmx_vmread, vmx_vmwrite},
};
use crate::arch::kvm::vmx::mmu::VmcsFields::CTRL_EPTP_PTR;

// pub const PT64_ROOT_LEVEL: u32 = 4;
// pub const PT32_ROOT_LEVEL: u32 = 2;
// pub const PT32E_ROOT_LEVEL: u32 = 3;

// pub struct KvmMmuPage{
//     gfn: u64, // 管理地址范围的起始地址对应的 gfn
//     role: KvmMmuPageRole, // 基本信息，包括硬件特性和所属层级等
//     // spt: *mut u64, // spt: shadow page table,指向 struct page 的地址，其包含了所有页表项 (pte)。同时 page->private 会指向该 kvm_mmu_page
// }

#[bitfield(u32)]
pub struct KvmMmuPageRole {
    #[bits(4)]
    level: usize, // 页所处的层级
    cr4_pae: bool, // cr4.pae，1 表示使用 64bit gpte
    #[bits(2)]
    quadrant: usize, // 如果 cr4.pae=0，则 gpte 为 32bit，但 spte 为 64bit，因此需要用多个 spte 来表示一个 gpte，该字段指示是 gpte 的第几块
    direct: bool,
    #[bits(3)]
    access: usize, // 访问权限
    invalid: bool,        // 失效，一旦 unpin 就会被销毁
    nxe: bool,            // efer.nxe，不可执行
    cr0_wp: bool,         // cr0.wp, 写保护
    smep_andnot_wp: bool, // smep && !cr0.wp，SMEP启用，用户模式代码将无法执行位于内核地址空间中的指令。
    smap_andnot_wp: bool, // smap && !cr0.wp
    #[bits(8)]
    unused: usize,
    #[bits(8)]
    smm: usize, // 1 表示处于 system management mode, 0 表示非 SMM
}

//  We don't want allocation failures within the mmu code, so we preallocate
// enough memory for a single page fault in a cache.
// pub struct KvmMmuMemoryCache {
//     num_objs: u32,
//     objs: [*mut u8; KVM_NR_MEM_OBJS as usize],
// }
pub type KvmMmuPageFaultHandler =
    fn(vcpu: &mut VmxVcpu, gpa: u64, error_code: u32, prefault: bool) -> Result<(), SystemError>;
#[derive(Default)]
pub struct KvmMmu {
    pub root_hpa: u64,
    pub root_level: u32,
    pub base_role: KvmMmuPageRole,
    // ...还有一些变量不知道用来做什么
    pub get_cr3: Option<fn(&VmxVcpu) -> u64>,
    pub set_eptp: Option<fn(u64) -> Result<(), SystemError>>,
    pub page_fault: Option<KvmMmuPageFaultHandler>,
    // get_pdptr: Option<fn(& VmxVcpu, index:u32) -> u64>, // Page Directory Pointer Table Register?暂时不知道和CR3的区别是什么
    // inject_page_fault: Option<fn(&mut VmxVcpu, fault: &X86Exception)>,
    // gva_to_gpa: Option<fn(&mut VmxVcpu, gva: u64, access: u32, exception: &X86Exception) -> u64>,
    // translate_gpa: Option<fn(&mut VmxVcpu, gpa: u64, access: u32, exception: &X86Exception) -> u64>,
    // sync_page: Option<fn(&mut VmxVcpu, &mut KvmMmuPage)>,
    // invlpg: Option<fn(&mut VmxVcpu, gva: u64)>, // invalid entry
    // update_pte: Option<fn(&mut VmxVcpu, sp: &KvmMmuPage, spte: u64, pte: u64)>,
}

impl core::fmt::Debug for KvmMmu {
    fn fmt(&self, f: &mut core::fmt::Formatter<'_>) -> core::fmt::Result {
        f.debug_struct("KvmMmu")
            .field("root_hpa", &self.root_hpa)
            .field("root_level", &self.root_level)
            .field("base_role", &self.base_role)
            .finish()
    }
}

fn tdp_get_cr3(_vcpu: &VmxVcpu) -> u64 {
    let guest_cr3 = vmx_vmread(VmcsFields::GUEST_CR3 as u32).expect("Failed to read eptp");
    return guest_cr3;
}

fn tdp_set_eptp(root_hpa: u64) -> Result<(), SystemError> {
    // 设置权限位，目前是写死的，可读可写可执行
    //  EPT paging-structure memory type: Uncacheable
    let mut eptp = 0x0_u64;
    // This value is 1 less than the EPT page-walk length.  3 means 4-level paging.
    eptp |= 0x3 << 3;
    eptp |= root_hpa & (PAGE_MASK as u64);
    vmx_vmwrite(CTRL_EPTP_PTR as u32, eptp)?;
    Ok(())
}

fn tdp_page_fault(
    vcpu: &mut VmxVcpu,
    gpa: u64,
    error_code: u32,
    prefault: bool,
) -> Result<(), SystemError> {
    debug!("tdp_page_fault");
    let gfn = gpa >> PAGE_SHIFT; // 物理地址右移12位得到物理页框号(相对于虚拟机而言)
                                 // 分配缓存池，为了避免在运行时分配空间失败，这里提前分配/填充足额的空间
    mmu_topup_memory_caches(vcpu)?;
    // TODO：获取gfn使用的level，处理hugepage的问题
    let level = 1; // 4KB page
                   // TODO: 快速处理由读写操作引起violation，即present同时有写权限的非mmio page fault
                   // fast_page_fault(vcpu, gpa, level, error_code)
                   // gfn->pfn
    let mut map_writable = false;
    let write = error_code & ((1_u32) << 1);
    let pfn = mmu_gfn_to_pfn_fast(vcpu, gpa, prefault, gfn, write == 0, &mut map_writable)?;
    // direct map就是映射ept页表的过程
    __direct_map(vcpu, gpa, write, map_writable, level, gfn, pfn, prefault)?;
    Ok(())
}

/*
 * Caculate mmu pages needed for kvm.
 */
// pub fn kvm_mmu_calculate_mmu_pages() -> u32 {
// 	let mut nr_mmu_pages:u32;
//     let mut nr_pages = 0;

//     let kvm = vm(0).unwrap();
//     for as_id in 0..KVM_ADDRESS_SPACE_NUM {
//         let slots = kvm.memslots[as_id];
//         for i in 0..KVM_MEM_SLOTS_NUM {
//             let memslot = slots.memslots[i as usize];
//             nr_pages += memslot.npages;
//         }
//     }

// 	nr_mmu_pages = (nr_pages as u32)* KVM_PERMILLE_MMU_PAGES / 1000;
// 	nr_mmu_pages = nr_mmu_pages.max(KVM_MIN_ALLOC_MMU_PAGES);
// 	return nr_mmu_pages;
// }

// pub fn kvm_mmu_change_mmu_pages(mut goal_nr_mmu_pages: u32){
//     let kvm = KVM();
//     // 释放多余的mmu page
//     if kvm.lock().arch.n_used_mmu_pages > goal_nr_mmu_pages {
//         while kvm.lock().arch.n_used_mmu_pages > goal_nr_mmu_pages {
//             if !prepare_zap_oldest_mmu_page() {
//                 break;
//             }
//         }
//         // kvm_mmu_commit_zap_page();
//         goal_nr_mmu_pages = kvm.lock().arch.n_used_mmu_pages;

//     }
//     kvm.lock().arch.n_max_mmu_pages = goal_nr_mmu_pages;
// }

// pub fn prepare_zap_oldest_mmu_page() -> bool {
//     return false;
// }

pub fn kvm_mmu_setup(vcpu: &Mutex<VmxVcpu>) {
    // TODO: init_kvm_softmmu(vcpu), init_kvm_nested_mmu(vcpu)
    init_kvm_tdp_mmu(vcpu);
}

pub fn kvm_vcpu_mtrr_init(_vcpu: &Mutex<VmxVcpu>) -> Result<(), SystemError> {
    check_ept_features()?;
    Ok(())
}

pub fn init_kvm_tdp_mmu(vcpu: &Mutex<VmxVcpu>) {
    let context = &mut vcpu.lock().mmu;
    context.page_fault = Some(tdp_page_fault);
    context.get_cr3 = Some(tdp_get_cr3);
    context.set_eptp = Some(tdp_set_eptp);
    // context.inject_page_fault = kvm_inject_page_fault; TODO: inject_page_fault
    // context.invlpg = nonpaging_invlpg;
    // context.sync_page = nonpaging_sync_page;
    // context.update_pte = nonpaging_update_pte;

    // TODO: gva to gpa in kvm
    // if !is_paging(vcpu) { // vcpu不分页
    //     context.gva_to_gpa = nonpaging_gva_to_gpa;
    // 	context.root_level = 0;
    // } else if (is_long_mode(vcpu)) {
    // 	context.gva_to_gpa = paging64_gva_to_gpa;
    // 	context.root_level = PT64_ROOT_LEVEL;
    // TODO:: different paging strategy
    // } else if (is_pae(vcpu)) {
    //     context.gva_to_gpa = paging64_gva_to_gpa;
    //     context.root_level = PT32E_ROOT_LEVEL;
    // } else {
    //     context.gva_to_gpa = paging32_gva_to_gpa;
    //     context.root_level = PT32_ROOT_LEVEL;
    // }
}

#[allow(clippy::too_many_arguments)]
pub fn __direct_map(
    vcpu: &mut VmxVcpu,
    gpa: u64,
    _write: u32,
    _map_writable: bool,
    _level: i32,
    _gfn: u64,
    pfn: u64,
    _prefault: bool,
) -> Result<u32, SystemError> {
    debug!("gpa={}, pfn={}, root_hpa={:x}", gpa, pfn, vcpu.mmu.root_hpa);
    // 判断vcpu.mmu.root_hpa是否有效
    if vcpu.mmu.root_hpa == 0 {
        return Err(SystemError::KVM_HVA_ERR_BAD);
    }
    // 把gpa映射到hpa
    let mut ept_mapper = EptMapper::lock();
    let page_flags = EntryFlags::from_prot_flags(ProtFlags::from_bits_truncate(0x7_u64), false);
    unsafe {
        assert!(ept_mapper.walk(gpa, pfn << PAGE_SHIFT, page_flags).is_ok());
    }
    return Ok(0);
}

pub fn mmu_gfn_to_pfn_fast(
    vcpu: &mut VmxVcpu,
    _gpa: u64,
    _prefault: bool,
    gfn: u64,
    write: bool,
    writable: &mut bool,
) -> Result<u64, SystemError> {
    let slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
    let pfn = __gfn_to_pfn(slot, gfn, false, write, writable)?;
    Ok(pfn)
}

// TODO: 添加cache
pub fn mmu_topup_memory_caches(_vcpu: &mut VmxVcpu) -> Result<(), SystemError> {
    // 如果 vcpu->arch.mmu_page_header_cache 不足，从 mmu_page_header_cache 中分配
    // pte_list_desc_cache 和 mmu_page_header_cache 两块全局 slab cache 在 kvm_mmu_module_init 中被创建
    // mmu_topup_memory_cache(vcpu.mmu_page_header_cache,
    //     mmu_page_header_cache, 4);
    Ok(())
}