lkml.org 
[lkml]   [2004]   [Jun]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH] cpumask 5/10 rewrite cpumask.h - single bitmap based implementation
William Lee Irwin III <wli@holomorphy.com> wrote:
>> _SC_NPROCESSOR_CONF is
>> unimplementable. NR_CPUS serves as an upper bound on the number of cpus
>> that may at some time be simultaneously present in the future.

On Fri, Jun 04, 2004 at 11:27:30AM -0700, Andrew Morton wrote:
> NR_CPUS is arguably the correct thing when it comes to copying per-cpu info
> to and from userspace.
> Sometimes userspace wants to know NR_CPUS. Sometimes it wants to know the
> index of the max possible CPU. Sometimes, perhaps the index of the max
> online CPU. Sometimes the max index of the CPUs upon which this task is
> eligible to run. Sometimes (lame) userspace may want to know, at compile
> time, the maximum number of CPUs which a Linux kernel will ever support.
> It's not completely trivial.
> Which of the above is _SC_NPROCESSOR_CONF supposed to return?

_SC_NPROCESSORS_CONF looks like a glibc extension, and if so it's
somewhat arbitrary. It's not documented in the manpage or the header's
comments. I presumed it was the largest number of cpus that could be
simultaneously online in the running kernel instance (which is NR_CPUS
in current kernels). If it's a standard it's likely to be very poorly-
defined. These things differ as the implementations start varying e.g.
for very sparse cpuid spaces, but not anything we handle now.

I'd have to write more stuff to try to find the rest of these things.
I'm not sure all of them are implementable.

And the luserspace code needs the following atop all that:

--- nr_cpus.c.orig2 2004-06-04 11:23:28.130800560 -0700
+++ nr_cpus.c 2004-06-04 11:27:12.785647832 -0700
@@ -10,28 +10,35 @@

int main(void)
{
+ int cpus = detect_nr_cpus();
printf("%d\n", detect_nr_cpus());
- return 0;
+ return cpus <= 0;
}

static int detect_nr_cpus(void)
{
unsigned long *cpus = malloc(sizeof(long));
size_t upper, middle, lower = sizeof(long);
+ int ret = -ENOMEM;

+ if (!cpus)
+ return -ENOMEM;
for (upper = lower; getaffinity(0, upper, cpus) < 0; upper *= 2) {
if (!realloc(cpus, 2*upper))
- return -ENOMEM;
+ goto out;
}
while (lower < upper - 1) {
middle = (lower + upper)/2;
if (!realloc(cpus, middle))
- return -ENOMEM;
+ goto out;
if (getaffinity(0, middle, cpus) < 0)
lower = middle;
else
upper = middle;
}
+ ret = CHAR_BIT*upper;
+out:
+ free(cpus);
return CHAR_BIT*upper;
}

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
\
 
 \ /
  Last update: 2005-03-22 14:03    [from the cache]
©2003-2011 Jasper Spaans