C++ 多线程死锁(引入lock函数)
- 2020 年 2 月 14 日
- 筆記
上一篇讲了互斥锁(传送门)的用法,解决了多线程共享资源可能会造成的一些问题,那么引入了锁以后,其实也难免会造成一些问题,比如说忘记unlock,或者有两个锁a和b,一个锁a在等待锁b的解锁,锁b在等待锁a的解锁,这些情况都会造成程序的死锁,比如下面这个例子:
#include <iostream> #include <thread> #include <mutex> void work1(std::mutex& mylock1, std::mutex& mylock2) { for (int i = 0; i < 100000; i++) { mylock1.lock(); mylock2.lock(); std::cout << "work1 : " << i << std::endl; mylock2.unlock(); mylock1.unlock(); } } void work2(std::mutex& mylock1, std::mutex& mylock2) { for (int i = 0; i < 100000; i++) { mylock2.lock(); mylock1.lock(); std::cout << "work2 : " << i << std::endl; mylock1.unlock(); mylock2.unlock(); } } int main() { std::mutex mylock1, mylock2; int ans = 0; std::thread t1(work1, std::ref(mylock1), std::ref(mylock2)); std::thread t2(work2, std::ref(mylock1), std::ref(mylock2)); t1.join(); t2.join(); return 0; }
由于交叉加锁,使得两个锁都在等待对方解锁而造成的死锁,运行结果如下图所示:

解决这个死锁的问题只是把加锁的顺序改过来就可以了,然后也可以用std::lock函数来创建多个互斥锁,用法也很简单,首先创建两个互斥锁lock1和lock2,那么std::lock(lock1,lock2)这句代码就相当于lock1.lock();lock2.lock();,最后不要忘了对两个锁的unlock,其实也可以搭配lock_guard()来使用,因为lock_guard内部就有析构函数来unlock,所以在lock_guard中引用std::adopt_lock参数(作用是告诉编译器我已经lock过了,不需要再重复lock了)就可以实现省去后面的unlock语句了。代码如下:
#include <iostream> #include <thread> #include <mutex> void work1(std::mutex& mylock1, std::mutex& mylock2) { for (int i = 0; i < 100000; i++) { std::lock(mylock1, mylock2); std::lock_guard<std::mutex> lock1(mylock1, std::adopt_lock); std::lock_guard<std::mutex> lock2(mylock2, std::adopt_lock); std::cout << "work1 : " << i << std::endl; } } void work2(std::mutex& mylock1, std::mutex& mylock2) { for (int i = 0; i < 100000; i++) { std::lock(mylock1, mylock2); std::lock_guard<std::mutex> lock1(mylock1, std::adopt_lock); std::lock_guard<std::mutex> lock2(mylock2, std::adopt_lock); std::cout << "work2 : " << i << std::endl; } } int main() { std::mutex mylock1, mylock2; int ans = 0; std::thread t1(work1, std::ref(mylock1), std::ref(mylock2)); std::thread t2(work2, std::ref(mylock1), std::ref(mylock2)); t1.join(); t2.join(); return 0; }