Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!ames!think!kulla!barmar
From: barmar@kulla (Barry Margolin)
Newsgroups: comp.sys.mac
Subject: Re: PMMU...
Message-ID: <30153@news.Think.COM>
Date: 27 Sep 89 00:13:52 GMT
References: <32679@srcsip.UUCP> <35041@apple.Apple.COM>
Sender: news@Think.COM
Distribution: usa
Organization: Thinking Machines Corporation, Cambridge MA, USA
Lines: 23

In article <35041@apple.Apple.COM> chuq@Apple.COM (Chuq Von Rospach) writes:
>>>On a machine with memory management and
>>>a solid operating system, an application can't crash the system no matter
>>>what it does.  Essentially all workstation-class machines (Suns, Apollos,
>Um, that's a nice theoretical ideal, but I spent enough time at Sun crashing
>machines to know that applications *can* crash the system, even when they're
>not supposed to be able to. 

The operative words were "a solid operating system".  Unless there are
CPU or MMU design errors, memory management and protected mode can
generally protect a system from crashing due to user code.  Most
software crashes on multiuser OSes are due to bugs in the kernel;
since the kernel often runs in unprotected mode, such bugs CAN crash
the system.  Of course, once you discover such kernel bugs, it is
often easy to write applications that invoke them (I had an Ultrix 2.2
program that accessed tape drives, and running it (without
recompiling) on Ultrix 2.4 (the pre-release designation for 3.0)
caused the system to crash, even though the new system was supposed to
be binary-compatible with the old one).
Barry Margolin, Thinking Machines Corp.

barmar@think.com
{uunet,harvard}!think!barmar