LinuxQuestions.org - Determine what CPU my thread is on

- Programming (https://www.linuxquestions.org/questions/programming-9/)

- - Determine what CPU my thread is on (https://www.linuxquestions.org/questions/programming-9/determine-what-cpu-my-thread-is-on-817697/)

Determine what CPU my thread is on

Is there a library/system call that will return to me what CPU core a thread of execution is running on? I've looked for a bit on the net already and also in /usr/include and couldn't find one.

getcpu() and sched_getcpu() are two that I found, but when I include the appropriate header files (linux/getcpu.h, and sys/sched.h respectively), gcc says getcpu.h doesn't exist and the linker complains it cannot find the implementation in sched.h.

I'm sure I'm doing something really stupid or overthinking...

What I am doing is running an OpenMP application and specifying a list of cpu cores to run on with GOMP_CPU_AFFINITY. I want to make sure that each core is getting the same number of threads.

**UPDATE**

It seems that this code works, but only with glibc >= 2.6, and my machine has 2.5:

Code:

#define _GNU_SOURCE



#include <stdio.h>

#include <utmpx.h>



int main( void )

{

        printf( "cpu = %d\n", sched_getcpu() );

        return 0;

}

**UPDATE2**

It seems that getcpu() exists, but there is no glibc wrapper for it. The manpage for getcpu states that I must directly invoke it with syscall() and only with kernels 2.6.19+ (I have 2.6.27). Here's more of what I tried:

Code:

$ grep -rin SYS_getcpu /usr/include

/usr/include/bits/syscall.h:310:#define SYS_getcpu __NR_getcpu

$ grep -rin __NR_getcpu /usr/include

/usr/include/asm-i386/unistd.h:326:#define __NR_getcpu                318

/usr/include/bits/syscall.h:310:#define SYS_getcpu __NR_getcpu

$ cat testgetcpu.c

#define _GNU_SOURCE



#include <stdio.h>

#include <sys/syscall.h>

#include <unistd.h>

#include <errno.h>

#include <string.h>



int main( void )

{

        int cpu = syscall(318);

        if (cpu < 0)

        {

                printf( "Error: errno = %d\n", errno ); fflush(stdout);

                printf( "Error: errno: %s\n", strerror(errno) ); fflush(stdout);

                return -1;

        }

        printf( "cpu = %d\n", cpu );



        return 0;

}

$ gcc testgetcpu.c 

$ ./a.out 

Error: errno = 38

Error: errno: Function not implemented

Anyone have any ideas?

-AM

Would sched_getaffinity() help?

Cheers, Steve

Use containers (without CPU affinity) - let the scheduler worry about the details. Go find some (other) real work to do - for yourself I mean.
This is micro-management - if your manager did it to you you would (or should) complain like hell. Let algorithms handle that crap, not carbon based life forms (i.e. you).

Look up cgroups.

@stevexyz
I thought about that, but haven't given it a try yet. The benchmark I run spawns 24 threads using OpenMP and I use an environment variable to tell it to put each thread on its own core. I'm getting weird results and want to verify that it is doing what I am telling it to. Thinking about it now, that may be what it is using, sched_setaffinity().

@syg00
I am doing high-performance computing research and need to know for a fact that thread N is on CPU core N. So no, I cannot be ignorant about what the scheduler is doing. Pinning is required for me. I will take a look at cgroups, thanks for the suggestion.