备注

AI Translation Notice

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

• Source document: kernel/locking/rwsem.md

• Translation time: 2026-01-13 06:32:36

• Translation model: hunyuan-turbos-latest

Please report issues via Community Channel

RwSem Read-Write Semaphore

1. Introduction

RwSem (Read-Write Semaphore) is a sleepable read-write lock used to protect shared data in process context. Unlike the spinlock-based RwLock, RwSem will yield the CPU and enter sleep state when it fails to acquire the lock, rather than busy-waiting.

1.1 Applicable Scenarios

Feature	RwLock (spinlock)	RwSem (semaphore)
Context	Process / Interrupt	Process context only
Waiting Method	Busy-waiting (spinning)	Sleeping (scheduling)
Applicable Scenario	Short critical sections	Long critical sections
Interrupt Context	Available	Not available

Important: RwSem cannot be used in interrupt context as it may sleep.

2. Core Design

2.1 Lock State Representation

RwSem uses an atomic integer (AtomicUsize) to maintain the lock state, with efficient state management through bitfield encoding:

64位系统状态位布局:
+--------+--------------+------------+----------+------------------+
| Bit 63 |   Bit 62     |  Bit 61    | Bit 60   |   Bits 59..0     |
+--------+--------------+------------+----------+------------------+
| WRITER | UPGRADEABLE  |  BEING     | OVERFLOW |  READER COUNT    |
|        |   READER     | UPGRADED   | DETECT   |   (读者计数)      |
+--------+--------------+------------+----------+------------------+

Field	Position	Description
WRITER	Bit 63	Writer lock, 1 indicates a writer holds the lock
UPGRADEABLE_READER	Bit 62	Upgradeable reader lock, 1 indicates an upgradeable reader holds the lock
BEING_UPGRADED	Bit 61	Upgrade-in-progress flag, 1 indicates upgrading from upgradeable reader to writer
MAX_READER	Bit 60	Reader overflow detection bit, set when reader count reaches 2^60
READER_COUNT	Bits 59…0	Current number of active readers

const READER: usize = 1;
const WRITER: usize = 1 << (usize::BITS - 1);              // Bit 63
const UPGRADEABLE_READER: usize = 1 << (usize::BITS - 2);  // Bit 62
const BEING_UPGRADED: usize = 1 << (usize::BITS - 3);      // Bit 61
const MAX_READER: usize = 1 << (usize::BITS - 4);          // Bit 60

2.2 Three Lock Modes

RwSem supports three lock modes, providing different levels of access permissions:

1. Read Lock

   • Multiple readers can hold the read lock concurrently

   • Provides read-only access to data (&T)

   • Mutually exclusive with writers and upgrade-in-progress readers

2. Write Lock

   • Only one writer can hold the write lock

   • Provides mutable access to data (&mut T)

   • Mutually exclusive with all other lock modes

3. Upgradeable Read Lock

   • Only one upgradeable reader can hold this lock

   • Initially provides read-only access (&T)

   • Can be atomically upgraded to a write lock

   • Mutually exclusive with writers and other upgradeable readers, but can coexist with regular readers

2.3 Wait Queue Design

pub struct RwSem<T: ?Sized> {
    lock: AtomicUsize,      // 锁状态
    queue: WaitQueue,       // 单一等待队列
    val: UnsafeCell<T>,     // 被保护的数据
}

Design Features:

• Uses a single WaitQueue to manage all waiters (readers, writers, upgradeable readers)

• Leverages the atomic semantics of WaitQueue.wait_until() to ensure correctness

• Balances fairness and performance through different wake-up strategies

3. Lock Acquisition Mechanism

3.1 Read Lock Acquisition

pub fn read(&self) -> RwSemReadGuard<'_, T> {
    self.queue.wait_until(|| self.try_read())
}

pub fn try_read(&self) -> Option<RwSemReadGuard<'_, T>> {
    let lock = self.lock.fetch_add(READER, Acquire);
    if lock & (WRITER | BEING_UPGRADED | MAX_READER) == 0 {
        // 无写者、无升级中的可升级读者、未溢出 → 成功
        Some(RwSemReadGuard {
            inner: self,
            _nosend: PhantomData,
        })
    } else {
        // 有阻塞因素，回滚计数
        self.lock.fetch_sub(READER, Release);
        None
    }
}

Acquisition Conditions:

• No writer (WRITER bit is 0)

• No upgrade-in-progress reader (BEING_UPGRADED bit is 0)

• Reader count not overflowed (MAX_READER bit is 0)

Key Design:

• Optimistically increments reader count first (fetch_add)

• Then checks blocking conditions

• Rolls back count if failed (fetch_sub)

• This “increment then check” approach performs better in non-contention scenarios

3.2 Write Lock Acquisition

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

pub fn try_write(&self) -> Option<RwSemWriteGuard<'_, T>> {
    if self.lock.compare_exchange(0, WRITER, Acquire, Relaxed).is_ok() {
        Some(RwSemWriteGuard {
            inner: self,
            _nosend: PhantomData,
        })
    } else {
        None
    }
}

Acquisition Conditions:

• Lock state must be 0 (no readers, no writers, no upgradeable readers)

• Uses CAS operation to ensure atomicity

3.3 Upgradeable Read Lock Acquisition

pub fn upread(&self) -> RwSemUpgradeableGuard<'_, T> {
    self.queue.wait_until(|| self.try_upread())
}

pub fn try_upread(&self) -> Option<RwSemUpgradeableGuard<'_, T>> {
    let lock = self.lock.fetch_or(UPGRADEABLE_READER, Acquire)
                 & (WRITER | UPGRADEABLE_READER);
    if lock == 0 {
        // 无写者且无其他可升级读者 → 成功
        return Some(RwSemUpgradeableGuard {
            inner: self,
            _nosend: PhantomData,
        });
    } else if lock == WRITER {
        // 有写者，需要回滚
        self.lock.fetch_sub(UPGRADEABLE_READER, Release);
    }
    // lock == UPGRADEABLE_READER 表示已有其他可升级读者，
    // fetch_or 没有改变状态，无需回滚
    None
}

Acquisition Conditions:

• No writer (WRITER bit is 0)

• No other upgradeable readers (UPGRADEABLE_READER bit is 0)

• Can coexist with regular readers

Key Design:

• Uses fetch_or to atomically set the upgradeable reader bit

• Determines success by checking the returned old value

• Only needs to roll back if WRITER bit was set but failed

3.4 wait_until Core Mechanism

All blocking lock acquisitions rely on the WaitQueue.wait_until() method:

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. 创建一对 Waiter/Waker
    let (waiter, waker) = Waiter::new_pair();

    loop {
        // 3. 先注册 waker 到等待队列
        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）
    }
}

Key Points:

• Registers waker before checking conditions to avoid missing wake signals

• May fail to acquire the lock even after being woken (fair competition), requiring continued looping

• This design avoids complex wake-loss issues

4. Lock Release and Wake-up Strategies

4.1 Read Lock Release

impl<T: ?Sized> Drop for RwSemReadGuard<'_, T> {
    fn drop(&mut self) {
        // 原子递减读者计数
        if self.inner.lock.fetch_sub(READER, Release) == READER {
            // 这是最后一个读者，唤醒一个等待者
            self.inner.queue.wake_one();
        }
    }
}

Wake-up Strategy:

• Only wakes when the last reader releases

• Wakes one waiter (may be a writer or upgradeable reader)

• Avoids unnecessary wake-up overhead

4.2 Write Lock Release

impl<T: ?Sized> Drop for RwSemWriteGuard<'_, T> {
    fn drop(&mut self) {
        // 清除写者位
        self.inner.lock.fetch_and(!WRITER, Release);

        // 唤醒所有等待者
        self.inner.queue.wake_all();
    }
}

Wake-up Strategy:

• Wakes all waiters

• Allows multiple readers to acquire the lock concurrently

• Only one writer can succeed (through CAS competition)

4.3 Upgradeable Read Lock Release

impl<T: ?Sized> Drop for RwSemUpgradeableGuard<'_, T> {
    fn drop(&mut self) {
        let res = self.inner.lock.fetch_sub(UPGRADEABLE_READER, Release);
        if res == UPGRADEABLE_READER {
            // 没有其他读者，唤醒所有等待者
            self.inner.queue.wake_all();
        }
    }
}

Wake-up Strategy:

• If no other readers (res == UPGRADEABLE_READER), wakes all waiters

• If other readers remain, doesn’t wake (waits for last reader to wake)

5. Advanced Features

5.1 Write Lock Downgrade

Atomically downgrades a write lock to an upgradeable read lock, allowing concurrent reader access:

impl<'a, T> RwSemWriteGuard<'a, T> {
    pub fn downgrade(mut self) -> RwSemUpgradeableGuard<'a, T> {
        loop {
            self = match self.try_downgrade() {
                Ok(guard) => return guard,
                Err(e) => e,
            };
        }
    }

    fn try_downgrade(self) -> Result<RwSemUpgradeableGuard<'a, T>, Self> {
        let inner = self.inner;
        let res = self.inner.lock.compare_exchange(
            WRITER,
            UPGRADEABLE_READER,
            AcqRel,
            Relaxed,
        );
        if res.is_ok() {
            core::mem::forget(self);
            Ok(RwSemUpgradeableGuard {
                inner,
                _nosend: PhantomData,
            })
        } else {
            Err(self)
        }
    }
}

Use Cases:

• After writing data, needing to hold the lock for prolonged read operations

• Downgrading allows concurrent reader access, improving concurrency

Notes:

• Uses CAS operations to ensure atomicity during downgrade

• Uses loop retries to handle CAS failures (typically due to ABA issues)

• Doesn’t wake waiters after downgrade (handled during upgradeable read release)

5.2 Upgradeable Read Lock Upgrade

Atomically upgrades an upgradeable read lock to a write lock:

impl<'a, T> RwSemUpgradeableGuard<'a, T> {
    pub fn upgrade(mut self) -> RwSemWriteGuard<'a, T> {
        // 先设置升级标志，阻塞新的读者
        self.inner.lock.fetch_or(BEING_UPGRADED, Acquire);
        loop {
            self = match self.try_upgrade() {
                Ok(guard) => return guard,
                Err(e) => e,
            };
        }
    }

    pub fn try_upgrade(self) -> Result<RwSemWriteGuard<'a, T>, Self> {
        let res = self.inner.lock.compare_exchange(
            UPGRADEABLE_READER | BEING_UPGRADED,
            WRITER | UPGRADEABLE_READER,
            AcqRel,
            Relaxed,
        );
        if res.is_ok() {
            let inner = self.inner;
            core::mem::forget(self);
            Ok(RwSemWriteGuard {
                inner,
                _nosend: PhantomData,
            })
        } else {
            Err(self)
        }
    }
}

Upgrade Mechanism:

1. First sets BEING_UPGRADED flag to block new readers

2. Spins waiting for existing readers to release (reader count drops to 0)

3. Uses CAS to transition state from “upgradeable reader + upgrading” to “writer”

Key Design:

• Doesn’t sleep during upgrade, instead spins

• BEING_UPGRADED flag ensures no new readers enter

• Maintains UPGRADEABLE_READER bit until upgrade completes to prevent other threads from acquiring upgradeable read locks

5.3 Interruptible Lock Acquisition

Supports lock acquisition operations that can be interrupted by signals:

// 可中断的读锁获取
pub fn read_interruptible(&self) -> Result<RwSemReadGuard<'_, T>, SystemError> {
    self.queue.wait_until_interruptible(|| self.try_read())
}

// 可中断的写锁获取
pub fn write_interruptible(&self) -> Result<RwSemWriteGuard<'_, T>, SystemError> {
    self.queue.wait_until_interruptible(|| self.try_write())
}

Use Cases:

• User-space processes acquiring locks need to respond to signals (e.g., Ctrl+C)

• Prevents indefinite process blocking

Error Handling:

• Returns Err(SystemError::ERESTARTSYS) to indicate signal interruption

• Callers need to handle errors appropriately (typically returning to user-space)

6. API Reference

6.1 Creation

// 编译期常量初始化
pub const fn new(value: T) -> Self

// 运行时初始化
let rwsem = RwSem::new(data);

6.2 Read Lock Operations

// 阻塞获取（不可中断）
pub fn read(&self) -> RwSemReadGuard<'_, T>

// 阻塞获取（可被信号中断）
pub fn read_interruptible(&self) -> Result<RwSemReadGuard<'_, T>, SystemError>

// 非阻塞尝试获取
pub fn try_read(&self) -> Option<RwSemReadGuard<'_, T>>

6.3 Write Lock Operations

// 阻塞获取（不可中断）
pub fn write(&self) -> RwSemWriteGuard<'_, T>

// 阻塞获取（可被信号中断）
pub fn write_interruptible(&self) -> Result<RwSemWriteGuard<'_, T>, SystemError>

// 非阻塞尝试获取
pub fn try_write(&self) -> Option<RwSemWriteGuard<'_, T>>

6.4 Upgradeable Read Lock Operations

// 阻塞获取（不可中断）
pub fn upread(&self) -> RwSemUpgradeableGuard<'_, T>

// 非阻塞尝试获取
pub fn try_upread(&self) -> Option<RwSemUpgradeableGuard<'_, T>>

6.5 Lock Conversion Operations

impl<'a, T> RwSemWriteGuard<'a, T> {
    // 写锁降级为可升级读锁
    pub fn downgrade(self) -> RwSemUpgradeableGuard<'a, T>
}

impl<'a, T> RwSemUpgradeableGuard<'a, T> {
    // 可升级读锁升级为写锁
    pub fn upgrade(self) -> RwSemWriteGuard<'a, T>

    // 非阻塞尝试升级
    pub fn try_upgrade(self) -> Result<RwSemWriteGuard<'a, T>, Self>
}

6.6 Direct Access

// 获取可变引用（需要独占的 &mut self）
pub fn get_mut(&mut self) -> &mut T

7. Usage Examples

7.1 Basic Read-Write

use crate::libs::rwsem::RwSem;

static DATA: RwSem<Vec<u32>> = RwSem::new(Vec::new());

// 多个读者可并发访问
fn reader() {
    let guard = DATA.read();
    println!("Data: {:?}", *guard);
    // guard 离开作用域时自动释放读锁
}

// 写者独占访问
fn writer() {
    let mut guard = DATA.write();
    guard.push(42);
    // guard 离开作用域时自动释放写锁
}

7.2 Upgradeable Read Lock

fn reader_that_may_write() {
    // 先以可升级读者身份获取锁
    let guard = DATA.upread();

    // 读取数据判断是否需要修改
    if guard.is_empty() {
        // 需要修改，升级到写锁
        let mut write_guard = guard.upgrade();
        write_guard.push(1);
    }
    // 不需要修改，直接释放
}

7.3 Write Lock Downgrade

fn writer_with_downgrade() {
    // 先获取写锁进行修改
    let mut guard = DATA.write();
    guard.clear();
    guard.push(1);

    // 修改完成，降级为可升级读锁
    let read_guard = guard.downgrade();

    // 现在允许其他读者并发访问
    println!("After write: {:?}", *read_guard);
}

7.4 Interruptible Acquisition

fn interruptible_reader() -> Result<(), SystemError> {
    // 可被信号中断
    let guard = DATA.read_interruptible()?;
    println!("Data: {:?}", *guard);
    Ok(())
}

7.5 Non-blocking Attempt

fn try_reader() -> Option<()> {
    // 立即返回，不会睡眠
    let guard = DATA.try_read()?;
    println!("Data: {:?}", *guard);
    Some(())
}

8. Memory Ordering and Correctness

8.1 Memory Ordering Guarantees

RwSem uses the following memory orderings to ensure correctness:

• Acquire: Used when acquiring locks, ensures visibility of protected data

• Release: Used when releasing locks, ensures modifications in critical sections are visible to subsequent acquirers

• AcqRel: Operations requiring both Acquire and Release semantics (e.g., downgrades, upgrades)

• Relaxed: CAS failure paths, no synchronization needed

8.2 Happens-Before Relationships

写者释放 (Release) ────┐
                      │ happens-before
读者获取 (Acquire) ←──┘

Guarantees:

• All modifications by a writer before release are visible to subsequent readers

• No happens-before relationships between multiple readers (concurrent reads)

9. Comparison with Other Implementations

Feature	Linux rw_semaphore	Rust parking_lot::RwLock	DragonOS RwSem
State Storage	atomic_long_t	AtomicUsize	AtomicUsize
Wait Queue	Single queue + type tagging	Single queue + parking	Single queue + WaitQueue
Upgradeable Locks	Not supported	Supported	Supported
Lock Downgrade	Supported (down_write_to_read)	Supported	Supported
Fairness Policy	Writer priority + HANDOFF	FIFO + anti-starvation	FIFO fair competition
Interruptible Waits	Supported	Not supported	Supported
Interrupt Context	Not supported	Not supported	Not supported

10. Performance Characteristics

10.1 Fast Path Optimizations

• Uncontended Reads: Single atomic operation (fetch_add)

• Uncontended Writes: Single CAS operation

• Reader Release: Single atomic operation, only last reader wakes

10.2 Scalability

• Concurrent Reads: Fully concurrent among readers, no additional synchronization overhead

• Writer Wake-up: wake_all() allows concurrent reader wake-ups

• Queue Overhead: Only operates on the queue when waiters exist

10.3 Performance Recommendations

• Prefer try_* methods to avoid sleeping (in scenarios where quick retries are possible)

• Use upgradeable read locks to avoid deadlocks from read-to-write upgrades

• Outperforms in read-heavy scenarios compared to write-heavy scenarios

11. Notes

11.1 Usage Restrictions

1. Not for Interrupt Context - RwSem may sleep

2. Avoid Nested Locks - Recursive acquisition by the same thread leads to deadlocks

3. Avoid Read-to-Write Upgrades - Doesn’t support upgrading regular read locks to write locks (causes deadlocks)

4. Guards Not Thread-Safe - Guard types are marked as !Send, cannot be passed across threads

11.2 Deadlock Scenarios

// ❌ 错误：嵌套获取同一锁
let guard1 = rwsem.read();
let guard2 = rwsem.write();  // 死锁！

// ❌ 错误：尝试升级普通读锁
let guard = rwsem.read();
drop(guard);
let guard = rwsem.write();  // 不是原子升级，可能有竞态

// ✅ 正确：使用可升级读锁
let guard = rwsem.upread();
let guard = guard.upgrade();  // 原子升级

11.3 Best Practices

1. Prefer upgradeable read locks over regular read locks when writes might be needed

2. Use downgrades instead of release-and-reacquire (maintains atomicity)

3. Use *_interruptible variants in user-space processes

4. Minimize critical section sizes to avoid prolonged lock holding

12. Implementation Summary

                    ┌─────────────────────────────────────┐
                    │             RwSem<T>                │
                    │                                     │
                    │  ┌───────────────────────────────┐  │
                    │  │  lock: AtomicUsize            │  │
                    │  │  [W|U|B|O|READER_COUNT]       │  │
                    │  │  W: Writer                    │  │
                    │  │  U: Upgradeable Reader        │  │
                    │  │  B: Being Upgraded            │  │
                    │  │  O: Overflow Detect           │  │
                    │  └───────────────────────────────┘  │
                    │  ┌───────────────────────────────┐  │
                    │  │  queue: WaitQueue             │  │
                    │  │  (all waiters in FIFO order)  │  │
                    │  └───────────────────────────────┘  │
                    │  ┌───────────────────────────────┐  │
                    │  │  val: UnsafeCell<T>           │  │
                    │  └───────────────────────────────┘  │
                    └─────────────────────────────────────┘

锁获取流程（以读锁为例）:
    read() → wait_until(try_read)
                  │
                  ↓
           ┌──────────────┐
           │ 注册 waker    │ ← 先入队（防止唤醒丢失）
           └──────┬───────┘
                  │
                  ↓
           ┌──────────────┐
           │ 调用 cond()  │ ← 再检查条件
           │ = try_read() │
           └──────┬───────┘
                  │
            成功 ←┴─→ 失败
              │         │
              ↓         ↓
         返回 Guard   睡眠等待
                         │
                    被唤醒后循环

锁释放与唤醒:
    读锁: 最后一个读者 → wake_one()
    写锁: 总是 → wake_all()
    可升级读锁: 无其他读者时 → wake_all()

This design achieves an efficient, correct, and feature-rich read-write semaphore through clever use of bitfield encoding, the wait_until atomic waiting mechanism, and differentiated wake-up strategies.