misc/78569: seg fault compiling after install on AMD K6-2

[Dan Albers]

The following reply was made to PR misc/78569; it has been noted by GNATS.

From: Dan Albers <computing at kpfa.org>
To: Kris Kennaway <kris at freebsd.org>
Cc: freebsd-gnats-submit at freebsd.org
Subject: Re: misc/78569: seg fault compiling after install on AMD K6-2
Date: Tue, 8 Mar 2005 19:38:59 -0800

 On Tuesday 08 March 2005 19:17, you wrote:
 > On Tue, Mar 08, 2005 at 03:54:18PM -0800, Dan Albers wrote:
 > > > If it was idling all night, the CPU wasn't generating much heat.  The
 > > > proposed issue here is that the cooling system can't handle the *peak*
 > > > loads, i.e. when the CPU is working very hard -- like during a
 > > > buildworld.  I've seen *exactly* this problem occur on multiple amd64
 > > > k6-2 and k6-3 machines, and I'm sure you can find many other case
 > > > stories in the mailing list archives.
 > >
 > > Huh? This is not an amd64 bit processor. It is a uniprocessor AMD K6-2
 > > 550.
 >
 > Typo; omit the '64'.
 >
 > > Actually I have not checked the lists, only Google, and
 > > I found one in which someone made sure (as mentioned in UPDATING) to not
 > > try to build with optimization > 2. No mention of AMD however.
 > >
 > > Which mailing list should I check?
 >
 > questions, stable, etc.
 >
 > > > > BTW, I put the HD back in the Intel Celeron system, did a CVSUP,
 > > > > built world and GENERIC kernel OK. Put the drive back into the AMD
 > > > > system, and just to see what would happen did 'make buildworld'. This
 > > > > time it looks like it failed trying to build library libiberty.a with
 > > > > the message cc: internal error: illegal instruction (program ld)
 > > >
 > > > Probably compiled with an incompatible CPUTYPE.
 > >
 > > Do you mean that when I built the compiler on the Celeron system, that I
 > > cannot use the compiler on an AMD system? (they are both i386).
 > >
 > > Where can I find out which CPUTYPE the compiler is expecting/using?
 >
 > In /etc/make.conf and associated manpage.
 >
 > > > > I have had problems building since about 5.2.1 (I think 5.1 compiled
 > > > > ok) on other  AMD machines as well.
 > > >
 > > > On working hardware, a buildworld of known-good source (e.g. the same
 > > > release you're running) will not fail.
 > >
 > > Uh, oh. This is working hardware. And I'm assuming that the source which
 > > came with the CD image is good (compiles ok on the Celeron). A few months
 > > ago i was able to build a 4.10 system on an AMD system and build OK. No
 > > heat probs, no compiler probs.
 > >
 > > Anyway, this is important to me because we have a lot of AMDs which are
 > > not ready for the bonepile yet, and I would like to be able to install
 > > FreeBSD on them and use them for KDE workstations,  and be able to build
 > > on each one.
 > >
 > > I am willing to systematically try to find out why this is happening. I
 > > have coded C for about 15years of my life, so if need be I can poke
 > > around in the source. I'd rather not if I can modify some config file. If
 > > you do not have the time or interest, please refer me to someone else.
 >
 > Oh, I hadn't noticed you'd dropped the GNATS address from the CC line.
 > You should attach a copy of our correspondence as a followup to your
 > PR so that others can try and help you as well.
 >
 > Kris
 
 Thanks. I'll forward this to GNATS.
 
 BTW /etc/make.conf is vanilla: does not mention cputype or optimization flags. 
 I've mounted the HD in the Celeron system for deployment, but that is the 
 last one I have. Sure would like to get this working with the AMD K6-2 xxx's.
 
 Dan.