pub mod barrier; pub mod bump; use alloc::vec::Vec; use hashbrown::HashSet; use x86::time::rdtsc; use x86_64::registers::model_specific::EferFlags; use crate::driver::serial::serial8250::send_to_default_serial8250_port; use crate::include::bindings::bindings::{ multiboot2_get_load_base, multiboot2_get_memory, multiboot2_iter, multiboot_mmap_entry_t, multiboot_tag_load_base_addr_t, }; use crate::libs::align::page_align_up; use crate::libs::lib_ui::screen_manager::scm_disable_put_to_window; use crate::libs::spinlock::SpinLock; use crate::mm::allocator::page_frame::{FrameAllocator, PageFrameCount, PageFrameUsage}; use crate::mm::memblock::mem_block_manager; use crate::{ arch::MMArch, mm::allocator::{buddy::BuddyAllocator, bump::BumpAllocator}, }; use crate::mm::kernel_mapper::KernelMapper; use crate::mm::page::{PageEntry, PageFlags, PAGE_1G_SHIFT}; use crate::mm::{MemoryManagementArch, PageTableKind, PhysAddr, VirtAddr}; use crate::{kdebug, kinfo, kwarn}; use system_error::SystemError; use core::arch::asm; use core::ffi::c_void; use core::fmt::Debug; use core::mem::{self}; use core::sync::atomic::{compiler_fence, AtomicBool, Ordering}; use super::kvm::vmx::vmcs::VmcsFields; use super::kvm::vmx::vmx_asm_wrapper::vmx_vmread; pub type PageMapper = crate::mm::page::PageMapper; /// 初始的CR3寄存器的值,用于内存管理初始化时,创建的第一个内核页表的位置 static mut INITIAL_CR3_VALUE: PhysAddr = PhysAddr::new(0); /// 内核的第一个页表在pml4中的索引 /// 顶级页表的[256, 512)项是内核的页表 static KERNEL_PML4E_NO: usize = (X86_64MMArch::PHYS_OFFSET & ((1 << 48) - 1)) >> 39; static INNER_ALLOCATOR: SpinLock>> = SpinLock::new(None); #[derive(Clone, Copy, Debug)] pub struct X86_64MMBootstrapInfo { kernel_load_base_paddr: usize, kernel_code_start: usize, kernel_code_end: usize, kernel_data_end: usize, kernel_rodata_end: usize, start_brk: usize, } pub(super) static mut BOOTSTRAP_MM_INFO: Option = None; /// @brief X86_64的内存管理架构结构体 #[derive(Debug, Clone, Copy, Hash)] pub struct X86_64MMArch; /// XD标志位是否被保留 static XD_RESERVED: AtomicBool = AtomicBool::new(false); impl MemoryManagementArch for X86_64MMArch { /// 4K页 const PAGE_SHIFT: usize = 12; /// 每个页表项占8字节,总共有512个页表项 const PAGE_ENTRY_SHIFT: usize = 9; /// 四级页表(PML4T、PDPT、PDT、PT) const PAGE_LEVELS: usize = 4; /// 页表项的有效位的index。在x86_64中,页表项的第[0, 47]位表示地址和flag, /// 第[48, 51]位表示保留。因此,有效位的index为52。 /// 请注意,第63位是XD位,表示是否允许执行。 const ENTRY_ADDRESS_SHIFT: usize = 52; const ENTRY_FLAG_DEFAULT_PAGE: usize = Self::ENTRY_FLAG_PRESENT; const ENTRY_FLAG_DEFAULT_TABLE: usize = Self::ENTRY_FLAG_PRESENT; const ENTRY_FLAG_PRESENT: usize = 1 << 0; const ENTRY_FLAG_READONLY: usize = 0; const ENTRY_FLAG_READWRITE: usize = 1 << 1; const ENTRY_FLAG_USER: usize = 1 << 2; const ENTRY_FLAG_WRITE_THROUGH: usize = 1 << 3; const ENTRY_FLAG_CACHE_DISABLE: usize = 1 << 4; const ENTRY_FLAG_NO_EXEC: usize = 1 << 63; /// x86_64不存在EXEC标志位,只有NO_EXEC(XD)标志位 const ENTRY_FLAG_EXEC: usize = 0; const ENTRY_FLAG_ACCESSED: usize = 0; const ENTRY_FLAG_DIRTY: usize = 0; /// 物理地址与虚拟地址的偏移量 /// 0xffff_8000_0000_0000 const PHYS_OFFSET: usize = Self::PAGE_NEGATIVE_MASK + (Self::PAGE_ADDRESS_SIZE >> 1); const KERNEL_LINK_OFFSET: usize = 0x100000; // 参考 https://code.dragonos.org.cn/xref/linux-6.1.9/arch/x86/include/asm/page_64_types.h#75 const USER_END_VADDR: VirtAddr = VirtAddr::new((Self::PAGE_ADDRESS_SIZE >> 1) - Self::PAGE_SIZE); const USER_BRK_START: VirtAddr = VirtAddr::new(0x700000000000); const USER_STACK_START: VirtAddr = VirtAddr::new(0x6ffff0a00000); const FIXMAP_START_VADDR: VirtAddr = VirtAddr::new(0xffffb00000000000); /// 设置FIXMAP区域大小为1M const FIXMAP_SIZE: usize = 256 * 4096; const MMIO_BASE: VirtAddr = VirtAddr::new(0xffffa10000000000); const MMIO_SIZE: usize = 1 << PAGE_1G_SHIFT; /// @brief 获取物理内存区域 unsafe fn init() { extern "C" { fn _text(); fn _etext(); fn _edata(); fn _erodata(); fn _end(); } Self::init_xd_rsvd(); let load_base_paddr = Self::get_load_base_paddr(); let bootstrap_info = X86_64MMBootstrapInfo { kernel_load_base_paddr: load_base_paddr.data(), kernel_code_start: _text as usize, kernel_code_end: _etext as usize, kernel_data_end: _edata as usize, kernel_rodata_end: _erodata as usize, start_brk: _end as usize, }; unsafe { BOOTSTRAP_MM_INFO = Some(bootstrap_info); } // 初始化物理内存区域(从multiboot2中获取) Self::init_memory_area_from_multiboot2().expect("init memory area failed"); kdebug!("bootstrap info: {:?}", unsafe { BOOTSTRAP_MM_INFO }); kdebug!("phys[0]=virt[0x{:x}]", unsafe { MMArch::phys_2_virt(PhysAddr::new(0)).unwrap().data() }); // 初始化内存管理器 unsafe { allocator_init() }; send_to_default_serial8250_port("x86 64 init done\n\0".as_bytes()); } /// @brief 刷新TLB中,关于指定虚拟地址的条目 unsafe fn invalidate_page(address: VirtAddr) { compiler_fence(Ordering::SeqCst); asm!("invlpg [{0}]", in(reg) address.data(), options(nostack, preserves_flags)); compiler_fence(Ordering::SeqCst); } /// @brief 刷新TLB中,所有的条目 unsafe fn invalidate_all() { compiler_fence(Ordering::SeqCst); // 通过设置cr3寄存器,来刷新整个TLB Self::set_table(PageTableKind::User, Self::table(PageTableKind::User)); compiler_fence(Ordering::SeqCst); } /// @brief 获取顶级页表的物理地址 unsafe fn table(table_kind: PageTableKind) -> PhysAddr { match table_kind { PageTableKind::Kernel | PageTableKind::User => { compiler_fence(Ordering::SeqCst); let cr3 = x86::controlregs::cr3() as usize; compiler_fence(Ordering::SeqCst); return PhysAddr::new(cr3); } PageTableKind::EPT => { let eptp = vmx_vmread(VmcsFields::CTRL_EPTP_PTR as u32).expect("Failed to read eptp"); return PhysAddr::new(eptp as usize); } } } /// @brief 设置顶级页表的物理地址到处理器中 unsafe fn set_table(_table_kind: PageTableKind, table: PhysAddr) { compiler_fence(Ordering::SeqCst); asm!("mov cr3, {}", in(reg) table.data(), options(nostack, preserves_flags)); compiler_fence(Ordering::SeqCst); } /// @brief 判断虚拟地址是否合法 fn virt_is_valid(virt: VirtAddr) -> bool { return virt.is_canonical(); } /// 获取内存管理初始化时,创建的第一个内核页表的地址 fn initial_page_table() -> PhysAddr { unsafe { return INITIAL_CR3_VALUE; } } /// @brief 创建新的顶层页表 /// /// 该函数会创建页表并复制内核的映射到新的页表中 /// /// @return 新的页表 fn setup_new_usermapper() -> Result { let new_umapper: crate::mm::page::PageMapper = unsafe { PageMapper::create(PageTableKind::User, LockedFrameAllocator) .ok_or(SystemError::ENOMEM)? }; let current_ktable: KernelMapper = KernelMapper::lock(); let copy_mapping = |pml4_entry_no| unsafe { let entry: PageEntry = current_ktable .table() .entry(pml4_entry_no) .unwrap_or_else(|| panic!("entry {} not found", pml4_entry_no)); new_umapper.table().set_entry(pml4_entry_no, entry) }; // 复制内核的映射 for pml4_entry_no in KERNEL_PML4E_NO..512 { copy_mapping(pml4_entry_no); } return Ok(crate::mm::ucontext::UserMapper::new(new_umapper)); } const PAGE_SIZE: usize = 1 << Self::PAGE_SHIFT; const PAGE_OFFSET_MASK: usize = Self::PAGE_SIZE - 1; const PAGE_MASK: usize = !(Self::PAGE_OFFSET_MASK); const PAGE_ADDRESS_SHIFT: usize = Self::PAGE_LEVELS * Self::PAGE_ENTRY_SHIFT + Self::PAGE_SHIFT; const PAGE_ADDRESS_SIZE: usize = 1 << Self::PAGE_ADDRESS_SHIFT; const PAGE_ADDRESS_MASK: usize = Self::PAGE_ADDRESS_SIZE - Self::PAGE_SIZE; const PAGE_ENTRY_SIZE: usize = 1 << (Self::PAGE_SHIFT - Self::PAGE_ENTRY_SHIFT); const PAGE_ENTRY_NUM: usize = 1 << Self::PAGE_ENTRY_SHIFT; const PAGE_ENTRY_MASK: usize = Self::PAGE_ENTRY_NUM - 1; const PAGE_NEGATIVE_MASK: usize = !((Self::PAGE_ADDRESS_SIZE) - 1); const ENTRY_ADDRESS_SIZE: usize = 1 << Self::ENTRY_ADDRESS_SHIFT; const ENTRY_ADDRESS_MASK: usize = Self::ENTRY_ADDRESS_SIZE - Self::PAGE_SIZE; const ENTRY_FLAGS_MASK: usize = !Self::ENTRY_ADDRESS_MASK; unsafe fn read(address: VirtAddr) -> T { return core::ptr::read(address.data() as *const T); } unsafe fn write(address: VirtAddr, value: T) { core::ptr::write(address.data() as *mut T, value); } unsafe fn write_bytes(address: VirtAddr, value: u8, count: usize) { core::ptr::write_bytes(address.data() as *mut u8, value, count); } unsafe fn phys_2_virt(phys: PhysAddr) -> Option { if let Some(vaddr) = phys.data().checked_add(Self::PHYS_OFFSET) { return Some(VirtAddr::new(vaddr)); } else { return None; } } unsafe fn virt_2_phys(virt: VirtAddr) -> Option { if let Some(paddr) = virt.data().checked_sub(Self::PHYS_OFFSET) { return Some(PhysAddr::new(paddr)); } else { return None; } } #[inline(always)] fn make_entry(paddr: PhysAddr, page_flags: usize) -> usize { return paddr.data() | page_flags; } } impl X86_64MMArch { unsafe fn get_load_base_paddr() -> PhysAddr { let mut mb2_lb_info: [multiboot_tag_load_base_addr_t; 512] = mem::zeroed(); send_to_default_serial8250_port("get_load_base_paddr begin\n\0".as_bytes()); let mut mb2_count: u32 = 0; multiboot2_iter( Some(multiboot2_get_load_base), &mut mb2_lb_info as *mut [multiboot_tag_load_base_addr_t; 512] as usize as *mut c_void, &mut mb2_count, ); if mb2_count == 0 { send_to_default_serial8250_port( "get_load_base_paddr mb2_count == 0, default to 1MB\n\0".as_bytes(), ); return PhysAddr::new(0x100000); } let phys = mb2_lb_info[0].load_base_addr as usize; return PhysAddr::new(phys); } unsafe fn init_memory_area_from_multiboot2() -> Result { // 这个数组用来存放内存区域的信息(从C获取) let mut mb2_mem_info: [multiboot_mmap_entry_t; 512] = mem::zeroed(); send_to_default_serial8250_port("init_memory_area_from_multiboot2 begin\n\0".as_bytes()); let mut mb2_count: u32 = 0; multiboot2_iter( Some(multiboot2_get_memory), &mut mb2_mem_info as *mut [multiboot_mmap_entry_t; 512] as usize as *mut c_void, &mut mb2_count, ); send_to_default_serial8250_port("init_memory_area_from_multiboot2 2\n\0".as_bytes()); let mb2_count = mb2_count as usize; let mut areas_count = 0usize; let mut total_mem_size = 0usize; for info_entry in mb2_mem_info.iter().take(mb2_count) { // Only use the memory area if its type is 1 (RAM) if info_entry.type_ == 1 { // Skip the memory area if its len is 0 if info_entry.len == 0 { continue; } total_mem_size += info_entry.len as usize; mem_block_manager() .add_block( PhysAddr::new(info_entry.addr as usize), info_entry.len as usize, ) .unwrap_or_else(|e| { kwarn!( "Failed to add memory block: base={:#x}, size={:#x}, error={:?}", info_entry.addr, info_entry.len, e ); }); areas_count += 1; } } send_to_default_serial8250_port("init_memory_area_from_multiboot2 end\n\0".as_bytes()); kinfo!("Total memory size: {} MB, total areas from multiboot2: {mb2_count}, valid areas: {areas_count}", total_mem_size / 1024 / 1024); return Ok(areas_count); } fn init_xd_rsvd() { // 读取ia32-EFER寄存器的值 let efer: EferFlags = x86_64::registers::model_specific::Efer::read(); if !efer.contains(EferFlags::NO_EXECUTE_ENABLE) { // NO_EXECUTE_ENABLE是false,那么就设置xd_reserved为true kdebug!("NO_EXECUTE_ENABLE is false, set XD_RESERVED to true"); XD_RESERVED.store(true, Ordering::Relaxed); } compiler_fence(Ordering::SeqCst); } /// 判断XD标志位是否被保留 pub fn is_xd_reserved() -> bool { // return XD_RESERVED.load(Ordering::Relaxed); // 由于暂时不支持execute disable,因此直接返回true // 不支持的原因是,目前好像没有能正确的设置page-level的xd位,会触发page fault return true; } } impl VirtAddr { /// @brief 判断虚拟地址是否合法 #[inline(always)] pub fn is_canonical(self) -> bool { let x = self.data() & X86_64MMArch::PHYS_OFFSET; // 如果x为0,说明虚拟地址的高位为0,是合法的用户地址 // 如果x为PHYS_OFFSET,说明虚拟地址的高位全为1,是合法的内核地址 return x == 0 || x == X86_64MMArch::PHYS_OFFSET; } } unsafe fn allocator_init() { let virt_offset = VirtAddr::new(page_align_up(BOOTSTRAP_MM_INFO.unwrap().start_brk)); let phy_offset = unsafe { MMArch::virt_2_phys(virt_offset) }.unwrap(); mem_block_manager() .reserve_block(PhysAddr::new(0), phy_offset.data()) .expect("Failed to reserve block"); let mut bump_allocator = BumpAllocator::::new(phy_offset.data()); kdebug!( "BumpAllocator created, offset={:?}", bump_allocator.offset() ); // 暂存初始在head.S中指定的页表的地址,后面再考虑是否需要把它加到buddy的可用空间里面! // 现在不加的原因是,我担心会有安全漏洞问题:这些初始的页表,位于内核的数据段。如果归还到buddy, // 可能会产生一定的安全风险(有的代码可能根据虚拟地址来进行安全校验) let _old_page_table = MMArch::table(PageTableKind::Kernel); let new_page_table: PhysAddr; // 使用bump分配器,把所有的内存页都映射到页表 { // 用bump allocator创建新的页表 let mut mapper: crate::mm::page::PageMapper> = crate::mm::page::PageMapper::::create( PageTableKind::Kernel, &mut bump_allocator, ) .expect("Failed to create page mapper"); new_page_table = mapper.table().phys(); kdebug!("PageMapper created"); // 取消最开始时候,在head.S中指定的映射(暂时不刷新TLB) { let table = mapper.table(); let empty_entry = PageEntry::::from_usize(0); for i in 0..MMArch::PAGE_ENTRY_NUM { table .set_entry(i, empty_entry) .expect("Failed to empty page table entry"); } } kdebug!("Successfully emptied page table"); let total_num = mem_block_manager().total_initial_memory_regions(); for i in 0..total_num { let area = mem_block_manager().get_initial_memory_region(i).unwrap(); // kdebug!("area: base={:?}, size={:#x}, end={:?}", area.base, area.size, area.base + area.size); for i in 0..((area.size + MMArch::PAGE_SIZE - 1) / MMArch::PAGE_SIZE) { let paddr = area.base.add(i * MMArch::PAGE_SIZE); let vaddr = unsafe { MMArch::phys_2_virt(paddr) }.unwrap(); let flags = kernel_page_flags::(vaddr); let flusher = mapper .map_phys(vaddr, paddr, flags) .expect("Failed to map frame"); // 暂时不刷新TLB flusher.ignore(); } } } unsafe { INITIAL_CR3_VALUE = new_page_table; } kdebug!( "After mapping all physical memory, DragonOS used: {} KB", bump_allocator.offset() / 1024 ); // 初始化buddy_allocator let buddy_allocator = unsafe { BuddyAllocator::::new(bump_allocator).unwrap() }; // 设置全局的页帧分配器 unsafe { set_inner_allocator(buddy_allocator) }; kinfo!("Successfully initialized buddy allocator"); // 关闭显示输出 scm_disable_put_to_window(); // make the new page table current { let mut binding = INNER_ALLOCATOR.lock(); let mut allocator_guard = binding.as_mut().unwrap(); kdebug!("To enable new page table."); compiler_fence(Ordering::SeqCst); let mapper = crate::mm::page::PageMapper::::new( PageTableKind::Kernel, new_page_table, &mut allocator_guard, ); compiler_fence(Ordering::SeqCst); mapper.make_current(); compiler_fence(Ordering::SeqCst); kdebug!("New page table enabled"); } kdebug!("Successfully enabled new page table"); } #[no_mangle] pub extern "C" fn rs_test_buddy() { test_buddy(); } pub fn test_buddy() { // 申请内存然后写入数据然后free掉 // 总共申请200MB内存 const TOTAL_SIZE: usize = 200 * 1024 * 1024; for i in 0..10 { kdebug!("Test buddy, round: {i}"); // 存放申请的内存块 let mut v: Vec<(PhysAddr, PageFrameCount)> = Vec::with_capacity(60 * 1024); // 存放已经申请的内存块的地址(用于检查重复) let mut addr_set: HashSet = HashSet::new(); let mut allocated = 0usize; let mut free_count = 0usize; while allocated < TOTAL_SIZE { let mut random_size = 0u64; unsafe { x86::random::rdrand64(&mut random_size) }; // 一次最多申请4M random_size %= 1024 * 4096; if random_size == 0 { continue; } let random_size = core::cmp::min(page_align_up(random_size as usize), TOTAL_SIZE - allocated); let random_size = PageFrameCount::from_bytes(random_size.next_power_of_two()).unwrap(); // 获取帧 let (paddr, allocated_frame_count) = unsafe { LockedFrameAllocator.allocate(random_size).unwrap() }; assert!(allocated_frame_count.data().is_power_of_two()); assert!(paddr.data() % MMArch::PAGE_SIZE == 0); unsafe { assert!(MMArch::phys_2_virt(paddr) .as_ref() .unwrap() .check_aligned(allocated_frame_count.data() * MMArch::PAGE_SIZE)); } allocated += allocated_frame_count.data() * MMArch::PAGE_SIZE; v.push((paddr, allocated_frame_count)); assert!(addr_set.insert(paddr), "duplicate address: {:?}", paddr); // 写入数据 let vaddr = unsafe { MMArch::phys_2_virt(paddr).unwrap() }; let slice = unsafe { core::slice::from_raw_parts_mut( vaddr.data() as *mut u8, allocated_frame_count.data() * MMArch::PAGE_SIZE, ) }; for (i, item) in slice.iter_mut().enumerate() { *item = ((i + unsafe { rdtsc() } as usize) % 256) as u8; } // 随机释放一个内存块 if !v.is_empty() { let mut random_index = 0u64; unsafe { x86::random::rdrand64(&mut random_index) }; // 70%概率释放 if random_index % 10 > 7 { continue; } random_index %= v.len() as u64; let random_index = random_index as usize; let (paddr, allocated_frame_count) = v.remove(random_index); assert!(addr_set.remove(&paddr)); unsafe { LockedFrameAllocator.free(paddr, allocated_frame_count) }; free_count += allocated_frame_count.data() * MMArch::PAGE_SIZE; } } kdebug!( "Allocated {} MB memory, release: {} MB, no release: {} bytes", allocated / 1024 / 1024, free_count / 1024 / 1024, (allocated - free_count) ); kdebug!("Now, to release buddy memory"); // 释放所有的内存 for (paddr, allocated_frame_count) in v { unsafe { LockedFrameAllocator.free(paddr, allocated_frame_count) }; assert!(addr_set.remove(&paddr)); free_count += allocated_frame_count.data() * MMArch::PAGE_SIZE; } kdebug!("release done!, allocated: {allocated}, free_count: {free_count}"); } } /// 全局的页帧分配器 #[derive(Debug, Clone, Copy, Hash)] pub struct LockedFrameAllocator; impl FrameAllocator for LockedFrameAllocator { unsafe fn allocate(&mut self, mut count: PageFrameCount) -> Option<(PhysAddr, PageFrameCount)> { count = count.next_power_of_two(); if let Some(ref mut allocator) = *INNER_ALLOCATOR.lock_irqsave() { return allocator.allocate(count); } else { return None; } } unsafe fn free(&mut self, address: crate::mm::PhysAddr, count: PageFrameCount) { assert!(count.data().is_power_of_two()); if let Some(ref mut allocator) = *INNER_ALLOCATOR.lock_irqsave() { return allocator.free(address, count); } } unsafe fn usage(&self) -> PageFrameUsage { if let Some(ref mut allocator) = *INNER_ALLOCATOR.lock_irqsave() { return allocator.usage(); } else { panic!("usage error"); } } } /// 获取内核地址默认的页面标志 pub unsafe fn kernel_page_flags(virt: VirtAddr) -> PageFlags { let info: X86_64MMBootstrapInfo = BOOTSTRAP_MM_INFO.unwrap(); if virt.data() >= info.kernel_code_start && virt.data() < info.kernel_code_end { // Remap kernel code execute return PageFlags::new().set_execute(true).set_write(true); } else if virt.data() >= info.kernel_data_end && virt.data() < info.kernel_rodata_end { // Remap kernel rodata read only return PageFlags::new().set_execute(true); } else { return PageFlags::new().set_write(true).set_execute(true); } } unsafe fn set_inner_allocator(allocator: BuddyAllocator) { static FLAG: AtomicBool = AtomicBool::new(false); if FLAG .compare_exchange(false, true, Ordering::SeqCst, Ordering::SeqCst) .is_err() { panic!("Cannot set inner allocator twice!"); } *INNER_ALLOCATOR.lock() = Some(allocator); } /// 低地址重映射的管理器 /// /// 低地址重映射的管理器,在smp初始化完成之前,需要使用低地址的映射,因此需要在smp初始化完成之后,取消这一段映射 pub struct LowAddressRemapping; impl LowAddressRemapping { // 映射64M const REMAP_SIZE: usize = 64 * 1024 * 1024; pub unsafe fn remap_at_low_address(mapper: &mut PageMapper) { for i in 0..(Self::REMAP_SIZE / MMArch::PAGE_SIZE) { let paddr = PhysAddr::new(i * MMArch::PAGE_SIZE); let vaddr = VirtAddr::new(i * MMArch::PAGE_SIZE); let flags = kernel_page_flags::(vaddr); let flusher = mapper .map_phys(vaddr, paddr, flags) .expect("Failed to map frame"); // 暂时不刷新TLB flusher.ignore(); } } /// 取消低地址的映射 pub unsafe fn unmap_at_low_address(mapper: &mut PageMapper, flush: bool) { for i in 0..(Self::REMAP_SIZE / MMArch::PAGE_SIZE) { let vaddr = VirtAddr::new(i * MMArch::PAGE_SIZE); let (_, _, flusher) = mapper .unmap_phys(vaddr, true) .expect("Failed to unmap frame"); if !flush { flusher.ignore(); } } } }