向勇 陈渝 李国良 任炬
2024年春季
课程幻灯片列表
提纲
提高性能、简化开发、加强安全、支持数据持久保存、支持应用的灵活性，支持进程间交互，支持线程和协程
在多线程中支持对共享资源的同步互斥访问
git clone https://github.com/rcore-os/rCore-Tutorial-v3.git cd rCore-Tutorial-v3 git checkout ch8
包含了多个同步互斥相关的多线程应用程序
user/src/bin/ ├── mpsc_sem.rs # 基于信号量的生产者消费者问题 ├── phil_din_mutex.rs # 基于互斥锁的哲学家问题 ├── race_adder_*.rs # 各种方式的全局变量累加问题 ├── sync_sem.rs # 基于信号量的同步操作 ├── test_condvar.rs # 基于条件变量的同步操作
内核代码的主要改进部分
os/src/ ├── sync │ ├── condvar.rs //条件变量 │ ├── mod.rs │ ├── mutex.rs //互斥锁 │ ├── semaphore.rs //信号量 │ └── up.rs ├── syscall │ ├── sync.rs //增加了互斥锁、信号量和条件变量相关系统调用 ├── task │ ├── process.rs //进程控制块增加了互斥锁、信号量和条件变量 ├── timer.rs // 添加与定时相关的TimerCondVar类条件变量
比如执行哲学家问题的应用程序，展示了5个哲学家用5把叉子进行思考/进餐/休息的过程。
Rust user shell >> phil_din_mutex time cost = 7271 '-' -> THINKING; 'x' -> EATING; ' ' -> WAITING #0: ------- xxxxxxxx---------- xxxx----- xxxxxx--xxx #1: ---xxxxxx-- xxxxxxx---------- x---xxxxxx #2: ----- xx---------xx----xxxxxx------------ xxxx #3: -----xxxxxxxxxx------xxxxx-------- xxxxxx-- xxxxxxxxx #4: ------ x------ xxxxxx-- xxxxx------ xx #0: ------- xxxxxxxx---------- xxxx----- xxxxxx--xxx >>
race_adder.rs
A //全局变量 A=A+1 //多个线程对A进行累加
多个线程执行上述代码，真的会出现Race Condition（竞争条件）吗？
pub fn main() -> i32 { let start = get_time(); let mut v = Vec::new(); for _ in 0..THREAD_COUNT { v.push(thread_create(f as usize, 0) as usize); // f函数是线程主体 } let mut time_cost = Vec::new(); for tid in v.iter() { time_cost.push(waittid(*tid)); } println!("time cost is {}ms", get_time() - start); assert_eq!(unsafe { A }, PER_THREAD * THREAD_COUNT); //比较累计值A 0 }
全局变量累加问题的多线程应用 race_adder.rs
unsafe fn f() -> ! { let mut t = 2usize; for _ in 0..PER_THREAD { let a = &mut A as *mut usize; // “缓慢执行”A=A+1 let cur = a.read_volatile(); // “缓慢执行”A=A+1 for _ in 0..500 { t = t * t % 10007; } // 增加切换概率 a.write_volatile(cur + 1); // “缓慢执行”A=A+1 } exit(t as i32) }
>> race_adder time cost is 31ms Panicked at src/bin/race_adder.rs:40, assertion failed: `(left == right)` left: `15788`, right: `16000` [kernel] Aborted, SIGABRT=6
每次都会执行都会出现Race Condition（竞争条件）！
基于原子操作的全局变量累加问题的多线程应用 race_adder_atomic.rs
race_adder_atomic.rs
unsafe fn f() -> ! { for _ in 0..PER_THREAD { while OCCUPIED .compare_exchange(false, true, Ordering::Relaxed, Ordering::Relaxed) .is_err() { yield_(); } // 基于CAS操作的近似spin lock操作 let a = &mut A as *mut usize; // “缓慢执行”A=A+1 let cur = a.read_volatile(); // “缓慢执行”A=A+1 for _ in 0..500 { t = t * t % 10007; } // 增加切换概率 a.write_volatile(cur + 1); // “缓慢执行”A=A+1 OCCUPIED.store(false, Ordering::Relaxed); // unlock操作 } ...
>> race_adder_atomic time cost is 29ms >> race_adder_loop
可以看到，执行速度快，且正确。
基于互斥锁的多线程应用 race_adder_mutex_[spin|block]
race_adder_mutex_[spin|block]
unsafe fn f() -> ! { let mut t = 2usize; for _ in 0..PER_THREAD { mutex_lock(0); //lock(id) let a = &mut A as *mut usize; // “缓慢执行”A=A+1 let cur = a.read_volatile(); // “缓慢执行”A=A+1 for _ in 0..500 { t = t * t % 10007; } // 增加切换概率 a.write_volatile(cur + 1); // “缓慢执行”A=A+1 mutex_unlock(0); //unlock(id) } exit(t as i32) }
基于互斥锁的全局变量累加问题的多线程应用 race_adder_mutex_spin
race_adder_mutex_spin
>> race_adder_mutex_spin time cost is 249ms # 执行系统调用，且进行在就绪队列上的取出/插入/等待操作
基于互斥锁的全局变量累加问题的多线程应用 race_adder_mutex_block
race_adder_mutex_block
>> race_adder_mutex_blocking time cost is 919ms # 执行系统调用，且进行在就绪队列+等待队列上的取出/插入/等待操作
spin mutex和 block mutex 的核心数据结构（全局变量）： UPSafeCell
UPSafeCell
pub struct UPSafeCell<T> { //允许在单核上安全**使用可变全局变量** inner: RefCell<T>, //提供内部可变性和运行时借用检查 } unsafe impl<T> Sync for UPSafeCell<T> {} //声明支持全局变量安全地在线程间共享 impl<T> UPSafeCell<T> { pub unsafe fn new(value: T) -> Self { Self { inner: RefCell::new(value) } } pub fn exclusive_access(&self) -> RefMut<'_, T> { self.inner.borrow_mut() //得到它包裹的数据的独占访问权 } }
spin mutex和 block mutex 的核心数据结构
pub struct MutexSpin { locked: UPSafeCell<bool>, //locked是被UPSafeCell包裹的布尔全局变量 } pub struct MutexBlocking { inner: UPSafeCell<MutexBlockingInner>, } pub struct MutexBlockingInner { locked: bool, wait_queue: VecDeque<Arc<TaskControlBlock>>, //等待获取锁的线程等待队列 }
spin mutex的相关函数
pub trait Mutex: Sync + Send { //Send表示跨线程 move，Sync表示跨线程share data fn lock(&self); fn unlock(&self); } fn unlock(&self) { let mut locked = self.locked.exclusive_access(); //独占访问locked *locked = false; // }
impl Mutex for MutexSpin { fn lock(&self) { loop { let mut locked = self.locked.exclusive_access(); //独占访问locked if *locked { drop(locked); suspend_current_and_run_next(); //把当前线程放到就绪队列末尾 continue; } else { *locked = true; //得到锁了，可以继续进入临界区执行 return; ...
block mutex的相关函数
impl Mutex for MutexBlocking { fn lock(&self) { let mut mutex_inner = self.inner.exclusive_access(); //独占访问mutex_inner if mutex_inner.locked { //把当前线程挂到此lock相关的等待队列中 mutex_inner.wait_queue.push_back(current_task().unwrap()); drop(mutex_inner); //把当前线程从就绪队列中取出，设置为阻塞态，切换到另一就绪线程执行 block_current_and_run_next(); } else { mutex_inner.locked = true; //得到锁了，可以继续进入临界区执行 } }
fn unlock(&self) { let mut mutex_inner = self.inner.exclusive_access(); assert!(mutex_inner.locked); //从等待队列中取出线程，并放入到就绪队列中 if let Some(waking_task) = mutex_inner.wait_queue.pop_front() { add_task(waking_task); } else { mutex_inner.locked = false; //释放锁 } }
基于信号量的多线程应用 sync_sem
sync_sem
pub fn main() -> i32 { // create semaphores assert_eq!(semaphore_create(0) as usize, SEM_SYNC); // create threads let threads = vec![ thread_create(first as usize, 0), thread_create(second as usize, 0), ]; // wait for all threads to complete for thread in threads.iter() { waittid(*thread as usize); } ...
unsafe fn first() -> ! { sleep(10); println!("First work and wakeup Second"); semaphore_up(SEM_SYNC); exit(0) } unsafe fn second() -> ! { println!("Second want to continue,but need to wait first"); semaphore_down(SEM_SYNC); println!("Second can work now"); exit(0) }
>> sync_sem Second want to continue,but need to wait first First work and wakeup Second Second can work now sync_sem passed!
semaphore的核心数据结构
pub struct Semaphore { pub inner: UPSafeCell<SemaphoreInner>, //UPSafeCell包裹的内部可变结构 } pub struct SemaphoreInner { pub count: isize, //信号量的计数值 pub wait_queue: VecDeque<Arc<TaskControlBlock>>, //信号量的等待队列 }
semaphore的相关函数
pub fn down(&self) { let mut inner = self.inner.exclusive_access(); inner.count -= 1; //信号量的计数值减一 if inner.count < 0 { inner.wait_queue.push_back(current_task().unwrap()); //放入等待队列 drop(inner); //把当前线程从就绪队列中取出，设置为阻塞态，切换到另一就绪线程执行 block_current_and_run_next(); } }
pub fn up(&self) { let mut inner = self.inner.exclusive_access(); inner.count += 1;//信号量的计数值加一 if inner.count <= 0 { //从等待队列中取出线程，并放入到就绪队列中 if let Some(task) = inner.wait_queue.pop_front() { add_task(task); } } }
基于互斥锁和条件变量的多线程应用 test_condvar
test_condvar
pub fn main() -> i32 { // create condvar & mutex assert_eq!(condvar_create() as usize, CONDVAR_ID); assert_eq!(mutex_blocking_create() as usize, MUTEX_ID); // create threads let threads = vec![ thread_create(first as usize, 0), thread_create(second as usize, 0),]; // wait for all threads to complete for thread in threads.iter() { waittid(*thread as usize); } ...
unsafe fn second() -> ! { println!("Second want to continue,but need to wait A=1"); mutex_lock(MUTEX_ID); while A == 0 { println!("Second: A is {}", A); condvar_wait(CONDVAR_ID, MUTEX_ID); } mutex_unlock(MUTEX_ID); println!("A is {}, Second can work now", A); exit(0) }
unsafe fn first() -> ! { sleep(10); println!("First work, Change A --> 1 and wakeup Second"); mutex_lock(MUTEX_ID); A = 1; condvar_signal(CONDVAR_ID); mutex_unlock(MUTEX_ID); exit(0) }
>> test_condvar Second: A is 0 First work, Change A --> 1 and wakeup Second A is 1, Second can work now
second
A==0
first
condvar的核心数据结构
pub struct Condvar { pub inner: UPSafeCell<CondvarInner>, //UPSafeCell包裹的内部可变结构 } pub struct CondvarInner { pub wait_queue: VecDeque<Arc<TaskControlBlock>>,//等待队列 }
condvar的相关函数
pub fn wait(&self, mutex: Arc<dyn Mutex>) { mutex.unlock(); //释放锁 let mut inner = self.inner.exclusive_access(); inner.wait_queue.push_back(current_task().unwrap()); //放入等待队列 drop(inner); //把当前线程从就绪队列中取出，设置为阻塞态，切换到另一就绪线程执行 block_current_and_run_next(); mutex.lock(); }
pub fn signal(&self) { let mut inner = self.inner.exclusive_access(); //从等待队列中取出线程，并放入到就绪队列中 if let Some(task) = inner.wait_queue.pop_front() { add_task(task); } }
sleep的设计实现
pub fn sys_sleep(ms: usize) -> isize { let expire_ms = get_time_ms() + ms; let task = current_task().unwrap(); add_timer(expire_ms, task); block_current_and_run_next(); 0 }
慈母龙英文名(maiasaura)的含义是“好妈妈蜥蜴”。1979年在美国蒙大拿，科学家们发现了一些恐龙窝，其中有小恐龙的骨架。于是他们把这种恐龙命名为慈母龙。。