Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!cmcl2!beta!hc!ames!sgi!wdl1!bobw
From: bobw@wdl1.UUCP (Robert Lee Wilson Jr.)
Newsgroups: comp.arch
Subject: Re: Demand paged virtual memory (was Re: Free Software Foundation ...)
Message-ID: <3490008@wdl1.UUCP>
Date: Thu, 8-Oct-87 16:36:27 EDT
Article-I.D.: wdl1.3490008
Posted: Thu Oct  8 16:36:27 1987
Date-Received: Sun, 11-Oct-87 10:35:30 EDT
References: <4558@oberon.USC.EDU>
Lines: 11

I think your instruction that checks validity will have to be a
special instruction that locks out interrupts for the subsequent
instruction (except for some special interrupts corresponding to
memory invalid): Otherwise you could get a context switch in there
with the result that memory checks valid on one instruction but by
the time you execute the next instruction (from the viewpoint of
this process) pages have been rearranged, etc., so that the real
memory reference faults, and you have the same restart/continue
problem that sparked this discussion in the first place!

(My opinions, not my employer's: As to facts, we're still arguing.)