Linux 中线程与 CPU 核的绑定

最近在对项目进行性能优化，由于在多核平台上，所以了解了些进程、线程绑定cpu核的问题，在这里将所学记录一下。

不管是线程还是进程，都是通过设置亲和性(affinity)来达到目的。对于进程的情况，一般是使用sched_setaffinity这个函数来实现，网上讲的也比较多，这里主要讲一下线程的情况。

与进程的情况相似，线程亲和性的设置和获取主要通过下面两个函数来实现：

 

 

 

 

 

 

 

int pthread_setaffinity_np(pthread_t thread, size_t cpusetsize，const cpu_set_t *cpuset);int pthread_getaffinity_np(pthread_t thread, size_t cpusetsize,cpu_set_t *cpuset);

1

2

3

4

int pthread_setaffinity_np(pthread_t thread, size_t cpusetsize，

const cpu_set_t *cpuset);

int pthread_getaffinity_np(pthread_t thread, size_t cpusetsize,

cpu_set_t *cpuset);

从函数名以及参数名都很明了，唯一需要点解释下的可能就是cpu_set_t这个结构体了。这个结构体的理解类似于select中的fd_set，可以理解为cpu集，也是通过约定好的宏来进行清除、设置以及判断：

 

 

 

 

 

 

 

//初始化，设为空void CPU_ZERO (cpu_set_t *set);//将某个cpu加入cpu集中void CPU_SET (int cpu, cpu_set_t *set);//将某个cpu从cpu集中移出void CPU_CLR (int cpu, cpu_set_t *set);//判断某个cpu是否已在cpu集中设置了int CPU_ISSET (int cpu, const cpu_set_t *set);

1

2

3

4

5

6

7

8

//初始化，设为空

void CPU_ZERO (cpu_set_t *set);

//将某个cpu加入cpu集中

void CPU_SET (int cpu, cpu_set_t *set);

//将某个cpu从cpu集中移出

void CPU_CLR (int cpu, cpu_set_t *set);

//判断某个cpu是否已在cpu集中设置了

int CPU_ISSET (int cpu, const cpu_set_t *set);

cpu集可以认为是一个掩码，每个设置的位都对应一个可以合法调度的 cpu，而未设置的位则对应一个不可调度的 CPU。换而言之，线程都被绑定了，只能在那些对应位被设置了的处理器上运行。通常，掩码中的所有位都被置位了，也就是可以在所有的cpu中调度.

 

以下为测试代码：

 

 

 

 

 

 

 

#define _GNU_SOURCE#include #include #include #include #include #include void *myfun(void *arg){ cpu_set_t mask; cpu_set_t get; char buf[256]; int i; int j; int num = sysconf(_SC_NPROCESSORS_CONF); printf("system has %d processor(s)\n", num); for (i = 0; i < num; i++) { CPU_ZERO(&mask); CPU_SET(i, &mask); if (pthread_setaffinity_np(pthread_self(), sizeof(mask), &mask) < 0) { fprintf(stderr, "set thread affinity failed\n"); } CPU_ZERO(&get); if (pthread_getaffinity_np(pthread_self(), sizeof(get), &get) < 0) { fprintf(stderr, "get thread affinity failed\n"); } for (j = 0; j < num; j++) { if (CPU_ISSET(j, &get)) { printf("thread %d is running in processor %d\n", (int)pthread_self(), j); } } j = 0; while (j++ < 100000000) { memset(buf, 0, sizeof(buf)); } } pthread_exit(NULL);}int main(int argc, char *argv[]){ pthread_t tid; if (pthread_create(&tid, NULL, (void *)myfun, NULL) != 0) { fprintf(stderr, "thread create failed\n"); return -1; } pthread_join(tid, NULL); return 0;}

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

#define _GNU_SOURCE

#include

#include

#include

#include

#include

#include

 

void *myfun(void *arg)

{

    cpu_set_t mask;

    cpu_set_t get;

    char buf[256];

    int i;

    int j;

    int num = sysconf(_SC_NPROCESSORS_CONF);

    printf("system has %d processor(s)\n", num);

 

    for (i = 0; i < num; i++) {

        CPU_ZERO(&mask);

        CPU_SET(i, &mask);

        if (pthread_setaffinity_np(pthread_self(), sizeof(mask), &mask) < 0) {

            fprintf(stderr, "set thread affinity failed\n");

        }

        CPU_ZERO(&get);

        if (pthread_getaffinity_np(pthread_self(), sizeof(get), &get) < 0) {

            fprintf(stderr, "get thread affinity failed\n");

        }

        for (j = 0; j < num; j++) {

            if (CPU_ISSET(j, &get)) {

                printf("thread %d is running in processor %d\n", (int)pthread_self(), j);

            }

        }

        j = 0;

        while (j++ < 100000000) {

            memset(buf, 0, sizeof(buf));

        }

    }

    pthread_exit(NULL);

}

 

int main(int argc, char *argv[])

{

    pthread_t tid;

    if (pthread_create(&tid, NULL, (void *)myfun, NULL) != 0) {

        fprintf(stderr, "thread create failed\n");

        return -1;

    }

    pthread_join(tid, NULL);

    return 0;

}

这段代码将使myfun线程在所有cpu中依次执行一段时间，在我的四核cpu上，执行结果为：

 

 

 

 

 

 

 

system has 4 processor(s) thread 1095604544 is running in processor 0 thread 1095604544 is running in processor 1 thread 1095604544 is running in processor 2 thread 1095604544 is running in processor 3

1

2

3

4

5

system has 4 processor(s)

thread 1095604544 is running in processor 0

thread 1095604544 is running in processor 1

thread 1095604544 is running in processor 2

thread 1095604544 is running in processor 3

在一些嵌入式设备中，运行的进程线程比较单一，如果指定进程线程运行于特定的cpu核，减少进程、线程的核间切换，有可能可以获得更高的性能。