Re: REGRESSION: Performance regressions from switching anon_vma->lock to mutex

[Andi Kleen <ak@linux.intel.com>]

[-- Attachment #1: Type: text/plain, Size: 856 bytes --]


> The fact is, glibc is just total crap.
>
> I tried to send uli a patch to just add caching. No go. I sent
> *another* patch to at least make glibc use a sane interface (and the
> cache if it needs to fall back on /proc/stat for some legacy reason).
> We'll see what happens.

FWIW a rerun with this modified LD_PRELOAD that does caching seems
to have the same performance as the version that does sched_getaffinity.

So you're right. Caching indeed helps and my assumption that the child
would only do it once was incorrect.

The only problem I see with it is that it doesn't handle CPU hotplug,
but Paul's suggestion would fix that too.

> Paul Eggbert suggested "caching for one second" - by just calling
> "gettimtofday()" to see how old the cache is. That would work too.
>

Maybe we need a "standard LD_PRELOAD library to improve glibc" @)

-Andi


[-- Attachment #2: sysconf-caching.c --]
[-- Type: text/plain, Size: 471 bytes --]

// gcc -fPIC -shared sysconf-caching.c -ldl  -o sysconf-caching.so
#define _GNU_SOURCE 1
#include <dlfcn.h>
#include <sched.h>
#include <unistd.h>

static long int (*real_sysconf)(int name);

long int sysconf(int name)
{
	if (!real_sysconf) 
		real_sysconf = dlsym(RTLD_NEXT, "sysconf");

	if (name == _SC_NPROCESSORS_ONLN) {
		static int cache = -1; 

		if (cache == -1)
			cache = real_sysconf( _SC_NPROCESSORS_ONLN);
		return cache;
	}

	return real_sysconf(name);
}