qiurui
/
DragonOS
mirror of https://github.com/DragonOS-Community/DragonOS.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
DragonOS/docs/locales/en/kernel/sched/wait_queue.md

29 KiB
Raw Blame History

:::{note} AI Translation Notice

This document was automatically translated by hunyuan-turbos-latest model, for reference only.

  • Source document: kernel/sched/wait_queue.md

  • Translation time: 2026-01-13 12:52:37

  • Translation model: hunyuan-turbos-latest

Please report issues via Community Channel

:::

DragonOS Wait Queue Mechanism

1. Overview

The DragonOS WaitQueue is a process synchronization primitive based on the Waiter/Waker pattern, used for process blocking and waking. Through atomic operations and a carefully designed wait-wake protocol, it completely solves the "wake loss" problem.

1.1 Core Features

  • Zero wake loss: Ensures no wake signals are missed through the "register before check" mechanism
  • Single-use Waiter/Waker: Creates a pair of objects for each wait to avoid state reuse issues
  • wait_until as core: Uses the wait_until API that returns resources as the foundation, with other APIs built upon it
  • Atomic wait and acquire: Supports atomically "waiting for condition and acquiring resource" (e.g., lock Guard)
  • Multi-core friendly: Wakers can be shared across CPUs, supporting concurrent wakeups
  • Signal awareness: Supports both interruptible and uninterruptible wait modes

2. Core Design

2.1 Waiter/Waker Pattern

The wait queue adopts a producer-consumer pattern, separating waiters and wakers:

pub struct Waiter {
    waker: Arc<Waker>,
    _nosend: PhantomData<Rc<()>>,  // 标记为 !Send
}

pub struct Waker {
    state: AtomicU8,          // 唤醒状态机
    target: Weak<PCB>,        // 目标进程
}

Key Features:

  • Waiter: Thread-local held (!Send/!Sync), can only wait on the thread that created it
  • Waker: Shared through Arc, can be passed and woken across CPUs/threads
  • Single-use design: Creates new Waiter/Waker pairs for each wait to avoid state pollution
  • State machine handshake: Idle/Sleeping/Notified/Closed four states ensure no wake loss during concurrent wakeups

2.2 Wait Queue Structure

pub struct WaitQueue {
    inner: SpinLock<InnerWaitQueue>,
    num_waiters: AtomicU32,  // 快速路径检查
}

struct InnerWaitQueue {
    dead: bool,                    // 队列失效标志
    waiters: VecDeque<Arc<Waker>>, // 等待者队列（FIFO）
}

Design Highlights:

  • Fast path optimization: num_waiters atomic counter allows lock-free checking if queue is empty
  • FIFO order: Uses VecDeque to ensure fairness
  • Death marker: dead flag for cleanup during resource destruction
  • Spinlock protection: Uses SpinLock instead of Mutex, avoiding recursive dependencies

3. Core API: wait_until Family

3.1 wait_until: Core Wait Primitive

wait_until is the core of the wait queue mechanism, with all other wait methods built upon it:

pub fn wait_until<F, R>(&self, cond: F) -> R
where
    F: FnMut() -> Option<R>,
{
    // 1. 快速路径：先检查条件
    if let Some(res) = cond() {
        return res;
    }

    // 2. 创建唯一的 Waiter/Waker 对
    let (waiter, waker) = Waiter::new_pair();

    loop {
        // 3. 先注册 waker 到队列（关键！）
        self.register_waker(waker.clone());

        // 4. 再检查条件（防止唤醒丢失）
        if let Some(res) = cond() {
            self.remove_waker(&waker);
            return res;
        }

        // 5. 睡眠等待被唤醒
        waiter.wait();

        // 6. 被唤醒后循环继续（可能是伪唤醒或竞争失败）
    }
}

Key Design Concepts:

  1. Enqueue before check: Ensures any wake between condition check and sleep is properly handled
  2. Returns resource rather than boolean: Option<R> rather than bool, allowing direct return of acquired resources (e.g., lock Guard)
  3. Loop retry: May fail competition after wakeup, requiring re-enqueue and recheck
  4. Single Waiter/Waker pair: Only creates one pair for entire wait process, simplifying lifecycle management

3.2 wait_until Variants

// 不可中断等待
pub fn wait_until<F, R>(&self, cond: F) -> R

// 可中断等待（可被信号中断）
pub fn wait_until_interruptible<F, R>(&self, cond: F)
    -> Result<R, SystemError>

// 带超时的可中断等待
pub fn wait_until_timeout<F, R>(&self, cond: F, timeout: Duration)
    -> Result<R, SystemError>

Return Value Semantics:

  • Ok(R): Condition met, returns acquired resource
  • Err(ERESTARTSYS): Interrupted by signal
  • Err(EAGAIN_OR_EWOULDBLOCK): Timeout

Cancellation Semantics:

  • Before returning due to timeout or signal interruption, close the corresponding Waker and recheck condition
  • This prevents wake loss due to interleaving of cancellation and concurrent wakeup

3.3 Why wait_until is Better than wait_event

Traditional wait_event series APIs only return a boolean:

// 传统方式（存在竞态）
wait_event(|| lock.is_available());
let guard = lock.acquire();  // ❌ 可能失败！另一个线程可能抢先获取

While wait_until can atomically "wait and acquire":

// wait_until 方式（无竞态）
let guard = wait_until(|| lock.try_acquire());  // ✅ 原子获取

Key Advantages:

  • Eliminates race window between "check and acquire"
  • More concise code with clearer semantics
  • Better performance (reduces one atomic operation)

4. Wait and Wake Flows

4.1 Detailed Wait Flow

┌─────────────────────┐
│  调用 wait_until    │
└──────────┬──────────┘
           │
           ↓
    ┌──────────────┐
    │ 快速路径：    │
    │ 检查条件      │
    └──────┬───────┘
           │
      满足? ────────┐
        │ 否       │ 是
        ↓          ↓
  ┌──────────┐  返回结果
  │ 创建      │
  │ Waiter/  │
  │ Waker    │
  └────┬─────┘
       │
       ↓ [循环开始]
  ┌──────────────┐
  │ 1. 注册 waker │ ← 关键：先入队
  │    到队列     │
  └──────┬───────┘
         │
         ↓
  ┌──────────────┐
  │ 2. 再次检查  │ ← 防止唤醒丢失
  │    条件       │
  └──────┬───────┘
         │
    满足? ────────┐
      │ 否       │ 是
      ↓          ↓
  ┌──────────┐ ┌──────────┐
  │ 检查信号 │ │ 移除      │
  │ (可中断) │ │ waker    │
  └────┬─────┘ └────┬─────┘
       │            │
  有信号? ──┐       ↓
    │ 否  │ 是   返回结果
    ↓     ↓
  ┌────┐ 返回
  │睡眠│ ERESTARTSYS
  └─┬──┘
    │
    ↓ [被唤醒]
  ┌──────────────┐
  │ 检查信号     │
  │ (可中断)     │
  └──────┬───────┘
         │
         ↓ [循环继续]

Key Points Explained:

  1. Enqueue before check:

    • If check before enqueue, wake signal may be lost if condition becomes true and wakeup occurs between false check return and enqueue
    • Enqueue first ensures any subsequent wake can find our waker

  2. Necessity of loop retry:

    • Wakeup doesn't guarantee condition is met (could be spurious wakeup)
    • Even if condition was met, it may become unmet again due to competition
    • Thus need to re-enqueue and recheck

  3. Advantages of single-use Waker:

    • Avoids complex lifecycle management
    • Uses state machine to consume notifications (Notified -> Idle), preventing duplicate delivery
    • Waker is re-registered in loop but uses same Arc<Waker>

4.2 Wake Flow

wake_one: Wake One Waiter

pub fn wake_one(&self) -> bool {
    // 快速路径：队列为空
    if self.is_empty() {
        return false;
    }

    loop {
        // 从队列头部取出一个 waker
        let next = {
            let mut guard = self.inner.lock_irqsave();
            let waker = guard.waiters.pop_front();
            if waker.is_some() {
                self.num_waiters.fetch_sub(1, Ordering::Release);
            }
            waker
        };

        let Some(waker) = next else { return false };

        // 释放锁后再唤醒（减少锁持有时间）
        if waker.wake() {
            return true;  // 成功唤醒
        }
        // waker 已失效（进程已退出），继续尝试下一个
    }
}

Design Points:

  • FIFO order: Takes from queue head to ensure fairness
  • Wake outside lock: Releases lock before calling waker.wake(), reducing lock contention
  • Auto skip invalid waker: If target process exited, automatically tries next

wake_all: Wake All Waiters

pub fn wake_all(&self) -> usize {
    // 快速路径：队列为空
    if self.is_empty() {
        return 0;
    }

    // 一次性取出所有 waker（减少锁持有时间）
    let wakers = {
        let mut guard = self.inner.lock_irqsave();
        let mut drained = VecDeque::new();
        mem::swap(&mut guard.waiters, &mut drained);
        self.num_waiters.store(0, Ordering::Release);
        drained
    };

    // 释放锁后逐个唤醒
    let mut woken = 0;
    for waker in wakers {
        if waker.wake() {
            woken += 1;
        }
    }
    woken
}

Design Points:

  • Batch take: Clears queue at once, minimizing lock hold time
  • Wake outside lock: Wakes individually after releasing lock, allowing immediate competition
  • Returns actual wake count: Distinguishes between "wake request" and "actual wake"

4.3 Waker Wake Mechanism

impl Waker {
    pub fn wake(&self) -> bool {
        // 原子状态机：Sleep 时真正唤醒；Idle 时只记账通知
        // Sleeping -> Notified: 触发 ProcessManager::wakeup
        // Idle -> Notified: 记录提前唤醒
    }

    pub fn close(&self) {
        // 关闭 waker，防止后续唤醒
        self.state.store(STATE_CLOSED, Ordering::Release);
    }

    fn consume_notification(&self) -> bool {
        // 消费通知（用于 block_current），Notified -> Idle
        ...
    }
}

Key Features:

  • Atomic state machine: state manages Idle/Sleeping/Notified/Closed
  • Weak reference to target process: Uses Weak<PCB> to avoid circular references, auto-cleanup when process exits
  • Memory ordering guarantees: Release/Acquire semantics ensure modifications before wake are visible to woken process

4.4 Blocking Current Process

fn block_current(waiter: &Waiter, interruptible: bool) -> Result<(), SystemError> {
    loop {
        // 快速路径：已被唤醒
        if waiter.waker.consume_notification() {
            return Ok(());
        }

        // 禁用中断，进入临界区
        let irq_guard = unsafe { CurrentIrqArch::save_and_disable_irq() };

        // 进入睡眠握手（处理"先唤后睡"竞态）
        match waiter.waker.prepare_sleep() {
            WakerSleepState::Notified | WakerSleepState::Closed => {
                drop(irq_guard);
                return Ok(());
            }
            WakerSleepState::Sleeping => {}
        }

        // 标记进程为睡眠状态
        ProcessManager::mark_sleep(interruptible)?;
        // 再次消费通知，避免 prepare_sleep 与 mark_sleep 的竞态
        if waiter.waker.consume_notification() {
            drop(irq_guard);
            return Ok(());
        }
        drop(irq_guard);  // 恢复中断

        // 调度到其他进程
        schedule(SchedMode::SM_NONE);

        // 被唤醒后，检查信号（可中断模式）
        if interruptible && Signal::signal_pending_state(...) {
            return Err(SystemError::ERESTARTSYS);
        }
    }
}

Key Design:

  • Handshake mechanism: prepare_sleep() first transitions state to Sleeping, then consumes notification after mark_sleep
  • Interrupt protection: mark_sleep must be executed with interrupts disabled
  • Signal check: After being scheduled back, checks for pending signals

5. Memory Ordering and Correctness

5.1 Memory Ordering Guarantees

The wait queue uses following memory orderings to ensure correctness:

Operation Memory Ordering Purpose
waker.wake() Release Ensures modifications before wake are visible to woken
waiter.wait() Acquire Ensures visibility of all modifications before wake
register_waker Release Ensures visibility of enqueue operations
num_waiters Acquire/Release Synchronizes counter modifications

5.2 Happens-Before Relationships

线程 A（唤醒方）              线程 B（等待方）
    │                           │
    │ 修改共享数据               │
    │                           │
    ↓                           ↓
wake() (Release)          register_waker()
    │                           │
    │ happens-before             │
    │ ─────────────────────────→ │
    │                           ↓
    │                      wait() (Acquire)
    │                           │
    │                           ↓
    │                      观察到共享数据修改

Guarantees:

  • All modifications by waker before calling wake() are visible to waiter
  • This forms the basis for correct synchronization

5.3 Race-Free Proof

Case 1: Enqueue before wake (normal flow)

时间轴：
T1: 等待方 register_waker()
T2: 等待方检查条件，返回 false
T3: 等待方准备睡眠
T4: 唤醒方 wake_one()          ← waker 在队列中，正常唤醒
T5: 等待方被唤醒

Case 2: Wake before enqueue (race to handle)

时间轴：
T1: 等待方检查条件，返回 false
T2: 唤醒方修改条件为 true
T3: 唤醒方 wake_one()          ← waker 还未入队，唤醒失败
T4: 等待方 register_waker()
T5: 等待方再次检查条件         ← 检测到 true，不会睡眠！

Key Point: Through "check after enqueue", even if wake occurs before enqueue, second check can detect condition is already met.

6. Compatible API: wait_event Family

For backward compatibility, provides wait_event series APIs implemented based on wait_until:

// 可中断等待，返回 Result<(), SystemError>
pub fn wait_event_interruptible<F, B>(
    &self,
    mut cond: F,
    before_sleep: Option<B>,
) -> Result<(), SystemError>
where
    F: FnMut() -> bool,
    B: FnMut(),
{
    self.wait_until_impl(
        || if cond() { Some(()) } else { None },
        true,
        None,
        before_sleep,
    )
}

// 不可中断等待
pub fn wait_event_uninterruptible<F, B>(
    &self,
    mut cond: F,
    before_sleep: Option<B>,
) -> Result<(), SystemError>
where
    F: FnMut() -> bool,
    B: FnMut(),
{
    self.wait_until_impl(
        || if cond() { Some(()) } else { None },
        false,
        None,
        before_sleep,
    )
}

// 带超时的可中断等待
pub fn wait_event_interruptible_timeout<F>(
    &self,
    mut cond: F,
    timeout: Option<Duration>,
) -> Result<(), SystemError>
where
    F: FnMut() -> bool,
{
    self.wait_until_impl(
        || if cond() { Some(()) } else { None },
        true,
        timeout,
        None::<fn()>,
    )
}

before_sleep Hook:

  • Executed after enqueue and before sleep
  • Typical use: release locks to avoid sleeping while holding them
  • Example: wait_event_interruptible(|| cond(), Some(|| drop(guard)))

7. Convenience Methods

7.1 Sleep and Release Lock

// 释放 SpinLock 并睡眠（可中断）
pub fn sleep_unlock_spinlock<T>(&self, to_unlock: SpinLockGuard<T>)
    -> Result<(), SystemError>

// 释放 Mutex 并睡眠（可中断）
pub fn sleep_unlock_mutex<T>(&self, to_unlock: MutexGuard<T>)
    -> Result<(), SystemError>

// 释放 SpinLock 并睡眠（不可中断）
pub fn sleep_uninterruptible_unlock_spinlock<T>(&self, to_unlock: SpinLockGuard<T>)

// 释放 Mutex 并睡眠（不可中断）
pub fn sleep_uninterruptible_unlock_mutex<T>(&self, to_unlock: MutexGuard<T>)

Usage Example:

let guard = lock.lock();
// 检查条件
if !condition_met() {
    // 需要等待，释放锁并睡眠
    wait_queue.sleep_unlock_spinlock(guard)?;
    // 被唤醒后需要重新获取锁
}

7.2 Queue Lifecycle Management

// 标记队列失效，唤醒并清空所有等待者
pub fn mark_dead(&self) {
    let mut drained = VecDeque::new();
    {
        let mut guard = self.inner.lock_irqsave();
        guard.dead = true;
        mem::swap(&mut guard.waiters, &mut drained);
        self.num_waiters.store(0, Ordering::Release);
    }
    for w in drained {
        w.wake();
        w.close();
    }
}

// 检查队列是否为空
pub fn is_empty(&self) -> bool {
    self.num_waiters.load(Ordering::Acquire) == 0
}

// 获取等待者数量
pub fn len(&self) -> usize {
    self.num_waiters.load(Ordering::Acquire) as usize
}

8. Event Wait Queue

Besides regular wait queues, also provides event mask-based wait queues:

pub struct EventWaitQueue {
    wait_list: SpinLock<Vec<(u64, Arc<Waker>)>>,
}

impl EventWaitQueue {
    // 等待特定事件
    pub fn sleep(&self, events: u64)

    // 等待特定事件并释放锁
    pub fn sleep_unlock_spinlock<T>(&self, events: u64, to_unlock: SpinLockGuard<T>)

    // 唤醒等待任意匹配事件的线程（位掩码 AND）
    pub fn wakeup_any(&self, events: u64) -> usize

    // 唤醒等待精确匹配事件的线程（相等比较）
    pub fn wakeup(&self, events: u64) -> usize

    // 唤醒所有等待者
    pub fn wakeup_all(&self)
}

Use Cases:

  • Waiting for multiple event types (e.g., READABLE | WRITABLE)
  • Waking specific waiters by event type
  • Example: socket poll/select implementation

9. Timeout Support

9.1 Timeout Mechanism

pub fn wait_until_timeout<F, R>(&self, cond: F, timeout: Duration)
    -> Result<R, SystemError>
where
    F: FnMut() -> Option<R>,
{
    self.wait_until_impl(cond, true, Some(timeout), None::<fn()>)
}

Implementation Principle:

  1. Calculate deadline: deadline = now + timeout
  2. Create timer that wakes waiter upon expiration
  3. After wakeup, check if timeout occurred:
    • If timer triggered, return EAGAIN_OR_EWOULDBLOCK
    • If condition met, cancel timer and return result

9.2 TimeoutWaker

pub struct TimeoutWaker {
    waker: Arc<Waker>,
}

impl TimerFunction for TimeoutWaker {
    fn run(&mut self) -> Result<(), SystemError> {
        // 定时器到期，唤醒等待者
        self.waker.wake();
        Ok(())
    }
}

Key Design:

  • Timer wakes through Waker::wake() rather than directly waking PCB
  • This allows Waiter::wait() to correctly observe wake state
  • Uses same mechanism as normal wake for consistency

10. Usage Examples

10.1 Semaphore Implementation

struct Semaphore {
    counter: AtomicU32,
    wait_queue: WaitQueue,
}

impl Semaphore {
    fn down(&self) -> Result<(), SystemError> {
        // 使用 wait_until 直接获取信号量
        self.wait_queue.wait_until(|| {
            let old = self.counter.load(Acquire);
            if old > 0 {
                // 尝试原子递减
                if self.counter.compare_exchange(
                    old, old - 1, Acquire, Relaxed
                ).is_ok() {
                    return Some(());  // 成功获取
                }
            }
            None  // 获取失败，继续等待
        });
        Ok(())
    }

    fn up(&self) {
        self.counter.fetch_add(1, Release);
        self.wait_queue.wake_one();
    }
}

Advantages:

  • Uses wait_until to ensure atomic "wait and acquire"
  • Avoids race window between "check and acquire"
  • Clean and concise code

10.2 Condition Variable Implementation

struct CondVar {
    wait_queue: WaitQueue,
}

impl CondVar {
    fn wait<T>(&self, guard: MutexGuard<T>) -> Result<MutexGuard<T>, SystemError> {
        let mutex = guard.mutex();

        // 使用 before_sleep 钩子释放锁
        let mut guard = Some(guard);
        self.wait_queue.wait_event_interruptible(
            || false,  // 等待 notify
            Some(|| {
                if let Some(g) = guard.take() {
                    drop(g);  // 释放锁
                }
            }),
        )?;

        // 重新获取锁
        mutex.lock_interruptible()
    }

    fn notify_one(&self) {
        self.wait_queue.wake_one();
    }

    fn notify_all(&self) {
        self.wait_queue.wake_all();
    }
}

10.3 RwSem Integration

impl<T: ?Sized> RwSem<T> {
    pub fn read(&self) -> RwSemReadGuard<'_, T> {
        // 直接返回 Guard，无竞态
        self.queue.wait_until(|| self.try_read())
    }

    pub fn write(&self) -> RwSemWriteGuard<'_, T> {
        self.queue.wait_until(|| self.try_write())
    }

    pub fn read_interruptible(&self) -> Result<RwSemReadGuard<'_, T>, SystemError> {
        self.queue.wait_until_interruptible(|| self.try_read())
    }
}

Why this design:

  • try_read() returns Option<Guard>, perfectly matching wait_until requirements
  • One line of code implements complete "wait and acquire" logic
  • Compiler ensures type safety, preventing forgotten checks

10.4 Timed Wait

fn wait_with_timeout(queue: &WaitQueue, timeout_ms: u64) -> Result<(), SystemError> {
    let timeout = Duration::from_millis(timeout_ms);
    queue.wait_event_interruptible_timeout(
        || condition_met(),
        Some(timeout),
    )
}

11. Performance Characteristics

11.1 Fast Path Optimization

  • No waiters: is_empty() requires only one atomic read, no lock contention
  • Quick check: wait_until first checks condition to avoid unnecessary enqueue
  • Wake outside lock: Wake operations occur after releasing queue lock, reducing lock hold time

11.2 Scalability

  • FIFO queue: Ensures fairness, prevents starvation
  • Batch wake: wake_all() takes all wakers at once, minimizing lock contention
  • Cross-CPU wake: Wakers can wake across different CPUs without lock synchronization

11.3 Memory Overhead

  • Waiter: Stack allocated, no heap involvement
  • Waker: Shared through Arc, one per waiter
  • Queue: Only occupies space when waiters exist, minimal overhead for empty queue

12. Comparison and Evolution

12.1 Comparison with Traditional Implementations

Feature Traditional wait_event DragonOS wait_until
API Return bool Option<R>
Atomic Acquire Not supported (manual) Native support
Wake Loss Requires careful handling Design guarantees none
Race Window Check-acquire has window No window
Code Complexity High (manual state management) Low (compiler guaranteed)
Performance May require multiple atomics Minimizes atomics

12.2 Design Evolution

Old Design (Problematic):

// ❌ 可能存在唤醒丢失
loop {
    if condition() {
        return;
    }
    register_waker();
    sleep();
}

Current Design (Correct):

// ✅ 无唤醒丢失
loop {
    register_waker();  // 先入队
    if condition() {   // 再检查
        remove_waker();
        return;
    }
    sleep();
}

13. Best Practices

13.1 Usage Recommendations

  1. Prefer wait_until: Compared to wait_event, it provides stronger guarantees and cleaner code
  2. Leverage Option return value: Directly return acquired resources to avoid secondary acquisition
  3. Properly use interruptible variants: User-space processes should use *_interruptible to avoid inability to terminate
  4. Properly handle timeouts: Distinguish between ERESTARTSYS (signal) and EAGAIN_OR_EWOULDBLOCK (timeout)

13.2 Pitfalls to Avoid

// ❌ 错误：分离检查和获取
if condition() {
    let guard = acquire();  // 可能失败！
}

// ✅ 正确：原子等待并获取
let guard = wait_until(|| try_acquire());

// ❌ 错误：忘记处理信号
let result = wait_queue.wait_event_interruptible(|| cond(), None);
// 未检查 result

// ✅ 正确：正确处理错误
let result = wait_queue.wait_event_interruptible(|| cond(), None)?;

13.3 Debugging Suggestions

  • Use wait_queue.len() to check waiter count
  • Monitor state state machine changes (via logs)
  • Check for processes blocked for extended periods (possible wake logic errors)

14. Implementation Principle Summary

                    ┌─────────────────────────────────────┐
                    │           WaitQueue                 │
                    │  ┌───────────────────────────────┐  │
                    │  │ inner: SpinLock<InnerQueue>  │  │
                    │  │   ├─ dead: bool              │  │
                    │  │   └─ waiters: VecDeque       │  │
                    │  └───────────────────────────────┘  │
                    │  ┌───────────────────────────────┐  │
                    │  │ num_waiters: AtomicU32       │  │
                    │  └───────────────────────────────┘  │
                    └─────────────────────────────────────┘
                                    │
                        ┌───────────┴───────────┐
                        ↓                       ↓
            ┌────────────────────┐  ┌────────────────────┐
            │      Waiter        │  │       Waker        │
            │  ┌──────────────┐  │  │  ┌──────────────┐  │
            │  │ waker: Arc   │──┼──┼─→│ state:      │  │
            │  │              │  │  │  │  AtomicBool  │  │
            │  └──────────────┘  │  │  └──────────────┘  │
            │  ┌──────────────┐  │  │  ┌──────────────┐  │
            │  │ _nosend      │  │  │  │ target:      │  │
            │  │  PhantomData │  │  │  │  Weak<PCB>   │  │
            │  └──────────────┘  │  │  └──────────────┘  │
            └────────────────────┘  └────────────────────┘
                 (!Send)                   (Send+Sync)

核心流程：
┌──────────────────────────────────────────────────────────┐
│                   wait_until(cond)                       │
│                                                          │
│  1. 快速路径: if cond() → return                        │
│                                                          │
│  2. 创建 (waiter, waker) 对                             │
│                                                          │
│  3. loop {                                               │
│       ├─ register_waker(waker)      ← 先入队            │
│       ├─ if cond() → return         ← 再检查            │
│       ├─ waiter.wait()              ← 睡眠等待          │
│       └─ [被唤醒，循环继续]                             │
│     }                                                    │
└──────────────────────────────────────────────────────────┘

唤醒策略：
┌────────────────────┐
│    wake_one()      │  → 取出队首 waker → 唤醒一个进程
└────────────────────┘

┌────────────────────┐
│    wake_all()      │  → 取出所有 waker → 唤醒所有进程
└────────────────────┘

内存序保证：
    唤醒方: wake() (Release)
                │
                │ happens-before
                ↓
    等待方: wait() (Acquire)

This design achieves a zero wake loss, high-performance, easy-to-use wait queue mechanism through the core mechanisms of "enqueue before check", single-use Waiter/Waker pattern, and atomic memory ordering guarantees, providing a solid foundation for various synchronization primitives in DragonOS.