Re: Making glibc-2.13 work on other CPUs

[JohnT <jrt at worldlinc dot net>]

Since your systems are evidently local-networked, why not compile on a 
pentium I system? Shouldn't the build system produce a library coded for 
your pentium I systems if configure, makefiles and gcc work right? At 
least make the cpu and march settings match! Maybe set both to i586 or 
i386. Compiling on an old clunker to/from a network disk would take a 
while, but it sounds like a spare system has plenty of time. Start a 
compile, go fishing for a weekend, come back and see if it's done.

John T


1wlcyk3odfe729o1un.hadvabvobs@antichef.net wrote:
> Hello,
>
> I use a "master" machine for daily work, and a number of old boxes for casual smaller, routine tasks. The "master" used to be a AMD K6-II, and the others Pentium (I). The Pentiums are diskless and boot via bootpd from the master. They have their bootp-lib directory mounted over on startup with that one from the master, i.e. Pentium and K6-II share the same libc. This worked very well so far.
>
> No I've replaced the K6-II master with a Pentium-III, and at the same time upgraded to a new glibc (2.13), self compiled on the Pentium-III. Now the fun starts: when the Pentiums mount the (newly built) glibc-2.13, they throw the towel with "Illegal Instructions". I searched around a bit and came to the conclusion that the configure process spotted correctly a Pentium-III and churned out code tailored to it - alas a few commands are incomprehensible to the old Pentium (I)s. Bummer.
>
> No - what can I do to get a glibc-2.13 compiled that has only code that can be digested by Pentium I and III alike? I already tried to configure with --with-cpu=i386, and additionally fed gcc a -march=i486, but to no avail, Pentium I still gets illegal Instructions.
>
> There *must* be an easy way to do this, since Mr. Debian&  friends never differentiate between Pentium I and III, so I guess they settle for the smallest common denominator. Unfortunately, unless one of the savants helps me out here, I can try combinations of compile-options until the cows come home ...
>
> Thanks in advance.
> Andreas