Relation between stack limit and threads

Question

1. What is the relationship between ulimit -s <value> and the stack size (at thread level) in the Linux implementation (or for that matter any OS)?

   Is <number of threads> * <each thread stack size> must be less than < stack size assigned by ulimit command> valid justification?

2. In the below program - each thread allocates char [PTHREAD_STACK_MIN] and 10 threads are created. But when the ulimit is set to 10 * PTHREAD_STACK_MIN, it does not coredump due to abort. For some random value of stacksize (much less than 10 * PTHREAD_STACK_MIN), it core dumps. Why so?

My Understanding is that stacksize represents the stack occupied by all the threads in summation for the process.

Thread Function

#include <cstdio>  
#include <error.h>  
#include <unistd.h>  
#include <sys/select.h>  
#include <sys/time.h>  
#include <sys/resource.h>  
using namespace std;        
#include <pthread.h>  
#include <bits/local_lim.h>  

const unsigned int nrOfThreads = 10;  
pthread_t  ntid[nrOfThreads];


void* thr_fn(void* argv)
{
    size_t _stackSz;  
    pthread_attr_t _attr;  
    int err;

    err = pthread_attr_getstacksize(&_attr,&_stackSz);
    if( 0 != err)
    {
        perror("pthread_getstacksize");
    }

    printf("Stack size - %lu, Thread ID - %llu, Process Id - %llu \n", static_cast<long unsigned int> (_stackSz), static_cast<long long unsigned int> (pthread_self()), static_cast<long long unsigned int> (getpid()) );


    //check the stack size by actual allocation - equal to 1 + PTHREAD_STACK_MIN
    char a[PTHREAD_STACK_MIN ] = {'0'};

    struct timeval tm;
    tm.tv_sec = 1;
    while (1)
        select(0,0,0,0,&tm);

    return ( (void*) NULL);
} 

Main Function

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

{  

    struct rlimit rlim;
    int err;

    err = getrlimit(RLIMIT_STACK,&rlim);
    if( 0 != err)
    {
        perror("pthread_create ");
        return -1;
    }

    printf("Stacksize hard limit - %ld, Softlimit - %ld\n", static_cast <long unsigned int> (rlim.rlim_max) , 
            static_cast <long unsigned int> (rlim.rlim_cur));

    for(unsigned int j = 0; j < nrOfThreads; j++)
    {
        err = pthread_create(&ntid[j],NULL,thr_fn,NULL);
        if( 0 != err)
        {
            perror("pthread_create ");
            return -1;
        }
    }

    for(unsigned int j = 0; j < nrOfThreads; j++)
    {
        err = pthread_join(ntid[j],NULL);
        if( 0 != err)
        {
            perror("pthread_join ");
            return -1;
        }
    }

    perror("Join thread success");

    return 0;
}

---

PS:
I am using Ubuntu 10.04 LTS version, with below specification.
Linux laptop 2.6.32-26-generic #48-Ubuntu SMP Wed Nov 24 10:14:11 UTC 2010 x86_64 GNU/Linux

Answer 1

On UNIX/Linux, getrlimit(RLIMIT_STACK) is only guaranteed to give the size of the main thread's stack. The OpenGroup's reference is explicit on that, "initial thread's stack":

http://www.opengroup.org/onlinepubs/009695399/functions/getrlimit.html

For Linux, there's a reference which indicates that RLIMIT_STACK is what will be used by default for any thread stack (for NPTL threading):

http://www.kernel.org/doc/man-pages/online/pages/man3/pthread_create.3.html

Generally, since the programmer can decide (by using nonstandard attributes when creating the thread) where to put the stack and/or how much stack to use for a new thread, there is no such thing as a "cumulative process stack limit". It rather comes out of the total RLIMIT_AS address space size.
But you do have a limit on the number of threads you can create,sysconf(PTHREAD_THREADS_MAX), and you do have a lower limit for the minimum size a thread stack must have,sysconf(PTHREAD_STACK_MIN).

Also, you can query the default stacksize for new threads:

pthread_attr_t attr;
size_t stacksize;
if (!pthread_attr_init(&attr) && !pthread_attr_getstacksize(&attr, &stacksize))
    printf("default stacksize for a new thread: %ld\n", stacksize);

I.e. default-initialize a set of pthread attributes and ask for what stacksize the system gave you.

Answer 2

In a threaded program, stacks for all threads (except the initial one) are allocated out of the heap, so RLIMIT_STACK has little or no relation to how much stack space you can use for your threads.