为了允许在线程或进程之间共享数据,同步时必须的,互斥锁和条件变量是同步的基本组成部分。
1、互斥锁
互斥锁是用来保护临界区资源,实际上保护的是临界区中被操纵的数据,互斥锁通常用于保护由多个线程或多进程分享的共享数据。一般是一些可供线程间使用的全局变量,来达到线程同步的目的,即保证任何时刻只有一个线程或进程在执行其中的代码。一般加锁的轮廓如下:
pthread_mutex_lock() 临界区 pthread_mutex_unlock()
互斥锁API
pthread_mutex_lock(pthread_mutex_t *mutex);
用此函数加锁时,如果mutex已经被锁住,当前尝试加锁的线程就会阻塞,直到互斥锁被其他线程释放。当此函数返回时,说明互斥锁已经被当前线程成功加锁.
pthread_mutex_trylock(pthread_mutex_t *mutex);
用此函数加锁时,如果mutex已经卑琐主,当前尝试加锁的线程不会阻塞,而是立即返回,返回的错误码为EBUSY,而不是阻塞等待。
pthread_mutex_unlock(pthread_mutex_t *mutex);
注意使用锁之前要记得初始化。互斥锁的初始化有两种初始化方式:
1.对于静态分配的互斥锁一半用宏赋值的方式初始化
eg: static pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
2.对于动态分配的互斥锁(如调用malloc)或分配在共享内存中,则必须调用pthread_mutex_init(pthread_mutex *mutex, pthread_mutexattr_t *mutexattr)函数来进行初始化。
例子1:写个程序实现生产者—消费者问题,先只考虑多个生产者线程之间的同步,直到所有的生产者线程都完成工作以后,才启动消费者线程。程序如下:
1 #include <stdio.h> 2 #include <stdlib.h> 3 #include <unistd.h> 4 #include <pthread.h> 5 #include <errno.h> 6 7 #define MAXNITEMS 1000000 8 #define MAXNTHREADS 100 9 10 int nitems; 11 12 struct 13 { 14 pthread_mutex_t mutex; 15 int buff[MAXNITEMS]; 16 int nput; 17 int nval; 18 } shared = { 19 PTHREAD_MUTEX_INITIALIZER 20 }; 21 22 void *produce(void*); 23 void *consume(void*); 24 25 int main(int argc,char *argv[]) 26 { 27 int i,nthreads,count[MAXNTHREADS]; 28 pthread_t tid_produce[MAXNTHREADS],tid_consume; 29 if(argc != 3) 30 { 31 printf("usage: producongs2 <#itmes> <#threads>.\n"); 32 exit(0); 33 } 34 nitems = atoi(argv[1]); 35 nthreads = atoi(argv[2]); 36 pthread_setconcurrency(nthreads); //设置线程并发级别 37 for(i=0;i<nthreads;++i) 38 { 39 count[i] = 0; 40 pthread_create(&tid_produce[i],NULL,produce,&count[i]); 41 } 42 for(i=0;i<nthreads;i++) 43 { 44 pthread_join(tid_produce[i],NULL); //等待线程退出 45 printf("count[%d] = %d\n",i,count[i]); 46 } 47 pthread_create(&tid_consume,NULL,consume,NULL); 48 pthread_join(tid_consume,NULL); //等待线程退出 49 exit(0); 50 } 51 52 void *produce(void *arg) 53 { 54 for(; ;) 55 { 56 pthread_mutex_lock(&shared.mutex); //加锁 57 if(shared.nput >= nitems) 58 { 59 pthread_mutex_unlock(&shared.mutex); //释放锁 60 return ; 61 } 62 shared.buff[shared.nput] = shared.nval; 63 shared.nput++; 64 shared.nval++; 65 pthread_mutex_unlock(&shared.mutex); //加锁 66 *((int*) arg) += 1; 67 } 68 } 69 void *consume(void *arg) 70 { 71 int i; 72 for(i=0;i<nitems;i++) 73 { 74 if(shared.buff[i] != i) 75 printf("buff[%d] = %d\n",i,shared.buff[i]); 76 } 77 return; 78 }
程序执行结果如下:
例子2:改进例子1,所有生产者线程启动后立即启动消费者线程,这样生产者线程产生数据的同时,消费者线程就能出来它,此时必须同步生产者和消费者,程序如下:
1 #include <stdio.h> 2 #include <stdlib.h> 3 #include <unistd.h> 4 #include <pthread.h> 5 #include <errno.h> 6 7 #define MAXNITEMS 1000000 8 #define MAXNTHREADS 100 9 10 int nitems; 11 12 struct 13 { 14 pthread_mutex_t mutex; 15 int buff[MAXNITEMS]; 16 int nput; 17 int nval; 18 } shared = { 19 PTHREAD_MUTEX_INITIALIZER 20 }; 21 22 void *produce(void*); 23 void *consume(void*); 24 void consume_wait(int); 25 int main(int argc,char *argv[]) 26 { 27 int i,nthreads,count[MAXNTHREADS]; 28 pthread_t tid_produce[MAXNTHREADS],tid_consume; 29 if(argc != 3) 30 { 31 printf("usage: producongs2 <#itmes> <#threads>.\n"); 32 exit(0); 33 } 34 nitems = atoi(argv[1]); 35 nthreads = atoi(argv[2]); 36 pthread_setconcurrency(nthreads+1); 37 //创建生产者线程 38 for(i=0;i<nthreads;++i) 39 { 40 count[i] = 0; 41 pthread_create(&tid_produce[i],NULL,produce,&count[i]); 42 } 43 //创建消费者线程 44 pthread_create(&tid_consume,NULL,consume,NULL); 45 for(i=0;i<nthreads;i++) 46 { 47 pthread_join(tid_produce[i],NULL); 48 printf("count[%d] = %d\n",i,count[i]); 49 } 50 //等待消费者线程退出 51 pthread_join(tid_consume,NULL); 52 exit(0); 53 } 54 55 void *produce(void *arg) 56 { 57 for(; ;) 58 { 59 pthread_mutex_lock(&shared.mutex); 60 if(shared.nput >= nitems) 61 { 62 pthread_mutex_unlock(&shared.mutex); 63 return ; 64 } 65 shared.buff[shared.nput] = shared.nval; 66 shared.nput++; 67 shared.nval++; 68 pthread_mutex_unlock(&shared.mutex); 69 *((int*) arg) += 1; 70 } 71 } 72 void *consume(void *arg) 73 { 74 int i; 75 for(i=0;i<nitems;i++) 76 { 77 consume_wait(i); 78 if(shared.buff[i] != i) 79 printf("buff[%d] = %d\n",i,shared.buff[i]); 80 } 81 return; 82 } 83 void consume_wait(int i) 84 { 85 for(; ;) //进行轮询,判断i是否已经由生产者生产 86 { 87 pthread_mutex_lock(&shared.mutex); 88 if(i<shared.nput) //i已经生产 89 { 90 pthread_mutex_unlock(&shared.mutex); 91 return; 92 } 93 pthread_mutex_unlock(&shared.mutex); 94 } 95 }
存在的问题:当消费者获取的条目尚没有准备好时,消费者线程一次次的循环去判断,每次给互斥锁解锁又上锁,这种轮询的办法浪费CPU时间。
2、条件变量
互斥锁用于上锁,条件变量用于等待,条件变量的使用是与互斥锁共通使用的。
2.1等待与信号发送
条件变量类型是pthread_cond_t,调用函数如下:
pthread_cond_wait(pthread_cond_t *cond, pthread_mutex_t *pmutex);
pthread_cond_signal(pthread_cond_t *pcond);
每个条件变量总是有一个互斥锁与之关联。现在采用条件变量实现生产者与消费者问题,程序如下:
1 #include <stdio.h> 2 #include <stdlib.h> 3 #include <unistd.h> 4 #include <pthread.h> 5 #include <errno.h> 6 7 #define MAXNITEMS 1000000 8 #define MAXNTHREADS 100 9 10 int nitems; 11 12 struct 13 { 14 pthread_mutex_t mutex; 15 int buff[MAXNITEMS]; 16 int nput; 17 int nval; 18 } shared = { 19 PTHREAD_MUTEX_INITIALIZER 20 }; 21 //条件变量 22 struct { 23 pthread_mutex_t mutex; 24 pthread_cond_t cond; 25 int nready; 26 }nready = { 27 PTHREAD_MUTEX_INITIALIZER,PTHREAD_COND_INITIALIZER 28 }; 29 30 void *produce(void*); 31 void *consume(void*); 32 33 int main(int argc,char *argv[]) 34 { 35 int i,nthreads,count[MAXNTHREADS]; 36 pthread_t tid_produce[MAXNTHREADS],tid_consume; 37 if(argc != 3) 38 { 39 printf("usage: producongs2 <#itmes> <#threads>.\n"); 40 exit(0); 41 } 42 nitems = atoi(argv[1]); 43 nthreads = atoi(argv[2]); 44 pthread_setconcurrency(nthreads+1); 45 for(i=0;i<nthreads;++i) 46 { 47 count[i] = 0; 48 pthread_create(&tid_produce[i],NULL,produce,&count[i]); 49 } 50 pthread_create(&tid_consume,NULL,consume,NULL); 51 for(i=0;i<nthreads;i++) 52 { 53 pthread_join(tid_produce[i],NULL); 54 printf("count[%d] = %d\n",i,count[i]); 55 } 56 pthread_join(tid_consume,NULL); 57 exit(0); 58 } 59 60 void *produce(void *arg) 61 { 62 printf("producer begins work\n"); 63 for(; ;) 64 { 65 pthread_mutex_lock(&shared.mutex); 66 if(shared.nput >= nitems) 67 { 68 pthread_mutex_unlock(&shared.mutex); 69 return ; 70 } 71 shared.buff[shared.nput] = shared.nval; 72 shared.nput++; 73 shared.nval++; 74 pthread_mutex_unlock(&shared.mutex); 75 pthread_mutex_lock(&nready.mutex); 76 if(nready.nready == 0) 77 pthread_cond_signal(&nready.cond); //通知消费者 78 nready.nready++; 79 pthread_mutex_unlock(&nready.mutex); 80 *((int*) arg) += 1; 81 } 82 } 83 void *consume(void *arg) 84 { 85 int i; 86 printf("consuemer begins work.\n"); 87 for(i=0;i<nitems;i++) 88 { 89 pthread_mutex_lock(&nready.mutex); 90 while(nready.nready == 0) 91 pthread_cond_wait(&nready.cond,&nready.mutex); //等待生产者 92 nready.nready--; 93 pthread_mutex_unlock(&nready.mutex); 94 if(shared.buff[i] != i) 95 printf("buff[%d] = %d\n",i,shared.buff[i]); 96 } 97 return; 98 }
程序执行结果如下:
总的来说,给条件变量发送信号的过程代码如下:
struct { pthread_mutex_t mutex; pthread_cond_t cond; //维护本条件的各个变量 }var = {PTHREAD_MUTEX_INITIALIZER,PTHREAD_COND_INITIALIZER,...} pthread_mutex_lock(&var.mutex); 设置条件为真 pthread_cond_signal(&var.cond); pthread_mutex_unlock(&var.mutex);
测试条件并进入睡眠以等待条件变为真的代码大体如下:
pthread_mutex_lock(&var.mutex); while(条件为假) pthread_cond_wait(&var.cond,&var.mutex); 修改条件 pthread_mutex_unlock(&var.mutex);
2.2定时等待和广播
通常pthread_cond_signal只是唤醒等待在相应条件变量上的一个线程,在某些情况下需要唤醒多个线程(例如读写者问题),可以调用pthread_cond_broadcast唤醒阻塞在相应条件变量上的所有线程。pthread_cond_timewait允许线程就阻塞时间设置一个限制值。API如下:
pthread_cond_broadcast(pthread_cond_t *cond);
pthread_cond_timedwait(pthread_cond_t *cond, pthread_mutex, const struct timespec *abstime);