Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!rutgers!ames!styx!ptsfa!ihnp4!chinet!steinmetz!davidsen
From: davidsen@steinmetz.steinmetz.UUCP (William E. Davidsen Jr)
Newsgroups: comp.sources.d,comp.unix.xenix,comp.arch,comp.unix.wizards
Subject: Re: ARC on XENIX V/286
Message-ID: <5901@steinmetz.steinmetz.UUCP>
Date: Wed, 29-Apr-87 12:24:03 EDT
Article-I.D.: steinmet.5901
Posted: Wed Apr 29 12:24:03 1987
Date-Received: Sat, 2-May-87 10:10:27 EDT
References: <259@devon.UUCP
Reply-To: davidsen@kbsvax.steinmetz.UUCP (William E. Davidsen Jr)
Organization: General Electric CRD, Schenectady, NY
Lines: 38
Keywords: ARC port XENIX
Xref: mnetor comp.sources.d:646 comp.unix.xenix:281 comp.arch:1142 comp.unix.wizards:2094

In article <259@devon.UUCP   paul@devon.UUCP (Paul Sutcliffe Jr.) writes:
  So far I have seen three different versions of ARC 5.12 posted to
  net.sources.  Each of these was supposedly working under some flavor of
  Un*x.  Someone even posted diffs to the original saying that the
  changes allowed ARC to run under XENIX V/286.

There is a bug in the version of ARC distributed, and I can tell you
what it is (but not where, yet). The version distributed failed on my
Xenix, too. I didn't much want to go after it with adb, but thought I'd
take it home and use sdb on a unix-pc. Lo and behold, the memory fault
went away and the program runs fine. Also ran on a SRV Vax.

Then I decided that it was the 80286 archetecture (again) and tried it
on Microport, where it worked, too. After MUCH digging I have located
the source of the problem. Machines which load all user data space into
one lump do memory management on the lump. On a Vax or 68k machine you
get no problem if an array index is off by one, as long as the array
isn't the last thing in memory.

Microport seems to do about the same thing, starting a new segment only
when really needed. Xenix large model seems to generate a large number
of segments, one for each array larger than (something). This give full
bounds checking on every array. The truth is that one of the array refs
is WRONG, and clobbers something else in user memory. The only way this
will fail is on hardware which does full bounds checking. I haven't
found the actual location of the problem yet, if someone beats me,
please post. I will certainly post it back to SEAware if I locate it
myself.

NOTE: I have cross posted this to several groups where there have been
either discussions of segmented machines or where the users are interested
in solving bizarre and obscure problems.

-- 
bill davidsen			sixhub \	ARPA: wedu@ge-crd.arpa
      ihnp4!seismo!rochester!steinmetz ->  crdos1!davidsen
				chinet /
"Stupidity, like virtue, is its own reward"