Re: Hardware or software?

[Mouse <mouse%Rodents-Montreal.ORG@localhost>]

>>> [year-old report of trouble with a 9000/712/80]
>> HP-PA, always nice...  [...]
> The frustrating part is that, based on the date of the cross-build, I
> feel reasonably confident it was built from my gitification of 5.2,
> so I should have the source tree in question as one of my commits
> around that time.  But I've been unable to find one that cross-builds
> and works.

Today I made a little more progress with this.

I found my backup disk from 2017-01-01, the oldest I still have that is
after the 2016-08-21 build date reported by that kernel.  I pulled off
the backup of the machine whose hostname appears in the build line.
The /usr/src there is identical to the last commit in my tree before
2016-08-21, the build directory implied by the kernel version line
exists, and the kernel it contains is identical to the working kernel.

I did notice that the machine was i386.  All my recent cross-builds
have been from amd64.  So I did a test-build from a modern 5.2 i386
machine.  That build fails to boot too, but it fails with a different
syndrome from the amd64 cross-build!  There obviously is a bug
somewhere, but I doubt it would be easy to find, nor that anyone but me
would care if I did.  Installing from the build artifacts there yields
a (superficially) working system.

I do note that the .tgz files in that directory are comparatively
large.  My recent i386-hosted cross-build is

-rw-r--r--  1 mouse    31687400 Aug 24 06:36 base.tgz
-rw-r--r--  1 mouse    46167919 Aug 24 06:42 comp.tgz
-rw-r--r--  1 mouse      362275 Aug 24 06:42 etc.tgz
-rw-r--r--  1 mouse     3360452 Aug 24 06:42 games.tgz
-rw-r--r--  1 mouse     3543118 Aug 24 06:26 kern-GENERIC.tgz
-rw-r--r--  1 mouse    12261035 Aug 24 06:43 man.tgz
-rw-r--r--  1 mouse     3493124 Aug 24 06:43 misc.tgz
-rw-r--r--  1 mouse     3752128 Aug 24 06:43 tests.tgz
-rw-r--r--  1 mouse     3460204 Aug 24 06:43 text.tgz

(the amd64-hosted one is similar) whereas the old one is

-rw-r--r--  1 mouse     255386418 Aug 21  2016 base.tgz
-rw-r--r--  1 mouse      48578329 Aug 21  2016 comp.tgz
-rw-r--r--  1 mouse        362243 Aug 21  2016 etc.tgz
-rw-r--r--  1 mouse      11904436 Aug 21  2016 games.tgz
-rw-r--r--  1 mouse       3542852 Aug 21  2016 kern-GENERIC.tgz
-rw-r--r--  1 mouse      12259495 Aug 21  2016 man.tgz
-rw-r--r--  1 mouse       3493187 Aug 21  2016 misc.tgz
-rw-r--r--  1 mouse      13121734 Aug 21  2016 tests.tgz
-rw-r--r--  1 mouse       6344100 Aug 21  2016 text.tgz

base, games, tests, and text are much bigger than their recent-build
counterparts.  On checking, I find that (for example) /bin/cat is
linked static in the old build but dynamic in the new.  This reinforces
my suspicion that the trouble is related to dynamic linking.

I'll be experimenting more - for example, seeing if I can coax my
modern 5.2 machines into (cross-)building a static system, or perhaps
if the 2016-built system I have is working enough to self-host a native
build.  I'd really like to end up with a working HP-PA machine!

If anyone recalls anything relevant from the 5.2 days, I'd be
interested.

/~\ The ASCII				  Mouse
\ / Ribbon Campaign
 X  Against HTML		mouse%rodents-montreal.org@localhost
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B