Re: [PATCH] cpumask 5/10 rewrite cpumask.h - single bitmap based implementation

[Paul Jackson <pj@sgi.com>]

William Lee Irwin III wrote:
> Some interface is needed to export NR_CPUS 

Well ... technically ... such an interface already exists.
However the word "interface" might be too kind a description.

/*
 * Ugly hack to get size of cpumask - keep calling sched_getaffinity
 * with masks of increasing size until it stops complaining -EINVAL
 * that the mask is too small.  Determines size of kernel cpumask,
 * in number of bytes.  Size will always be some multiple of the
 * size of an unsigned long.  Since it's ok to be too big, and since
 * kernel NR_CPUS is usually a power of two, just try each power
 * of two, until one works.  Contrary to the man page, a successful
 * sched_getaffinity() returns a positive number (the sizeof(cpumask_t),
 * to be exact), not zero.  Consider any return >= 0 to mean success.
 */

#include <errno.h>
extern int errno;

int cpumasksz()
{
	int nbytes;
	unsigned long *mask = 0;

	for (nbytes = sizeof(unsigned long); nbytes < 10000; nbytes *= 2) {
		int r;
		mask = (unsigned long *)realloc(mask, nbytes);
		if (mask == 0)
			return -ENOMEM;
		errno = 0;
		r = sched_getaffinity(0, nbytes, mask);
		if (r < 0 && errno == EINVAL)
			continue;
		if (r >= 0)
			break;
		return -errno;
	}
	free(mask);
	return nbytes;
}



-- 
                          I won't rest till it's the best ...
                          Programmer, Linux Scalability
                          Paul Jackson <pj@sgi.com> 1.650.933.1373