Linux emulation: -ffast-math causes floating point exception

[Johannes Hofmann <johannes.hofmann@xxxxxx>]

Some Linux binaries compiled with -ffast-math crash with floating
point exception. There also is a warning in dmesg:

FXRSTR: illegal FP MXCSR 00009fc0 didinit = 0

This is a bit unfortunate as for some reason the procps package
(pkill, pgrep, uptime, etc) is compiled with the -ffast-math switch
and the utilities crash.

The following mini-program shows the crash when compiled with
-ffast-math (as Linux binary):

#include <stdio.h>

int main(int argc, char **argv) {
        double d;
        
        scanf("%lf", &d);
        printf("%f\n", d);
}