NetNews Usenet Archive 1992 #16/NN_1992_16.iso/spool/comp/lang/c/11817

Internet Message Format | 1992-07-31 | 1.8 KB

Path: sparky!uunet!gatech!destroyer!gumby!yale!yale.edu!think.com!barmar
From: [e-mail] (Barry Margolin)
Newsgroups: comp.lang.c
Subject: Re: Pointers to freed memory
Date: 31 Jul 1992 15:43:34 GMT
Organization: Thinking Machines Corporation, Cambridge MA, USA
Lines: 27
Message-ID: <[e-mail]>
References: <[e-mail]> <[e-mail]> <[e-mail]>
NNTP-Posting-Host: telecaster.think.com

In article <[e-mail]> [e-mail] (Jos Horsmeier) writes:
>I never could have
>imagined a processor architecture where a validation on an address
>value would be done while just _loading_ that particular value. But
>such (ugly) animals seem to exist; I've never had the `pleasure'
>to meet one ...

I remember this coming up when we were considering porting Multics to a
(then) new Honeywell process (I think it was the DPS/88).  Multics uses
several invalid segments for various purposes; for instance, anything in
segment -1 is a null pointer.  Previous generations of hardware would trap
when dereferencing through such pointers, which is just what we wanted.
The new hardware trapped when loading addresses into pointer registers,
thus making our normal mechanism for loading null pointers invalid.  The
hardware designers probably thought they were doing a Good Thing by
checking pointers earlier on (I suppose it helps you find out where the
invalid pointer came from), but they never realized that some of us were
using invalid pointers for special purposes (I think we also expected to
run into some problems porting Lisp to this machine, since it stored its
tag fields in unused bits in pointer representations, and the new machine
may have checked that those were always 0).

--
Barry Margolin
System Manager, Thinking Machines Corp.

[e-mail]          {uunet,harvard}!think!barmar