Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!uwvax!astroatc!prairie!dan
From: dan@prairie.UUCP (Daniel M. Frank)
Newsgroups: comp.sys.intel
Subject: Re: Copy-on-write
Message-ID: <364@prairie.UUCP>
Date: Wed, 26-Nov-86 00:04:46 EST
Article-I.D.: prairie.364
Posted: Wed Nov 26 00:04:46 1986
Date-Received: Wed, 26-Nov-86 07:48:55 EST
References: <362@prairie.UUCP> <867@sfsup.UUCP> <363@prairie.UUCP>
Reply-To: dan@prairie.UUCP (Daniel M. Frank)
Distribution: na
Organization: Prairie Computing, Madison, Wisconsin
Lines: 31


   Yup, I was wrong about this.  I missed it because I was thinking in
terms of sharing the same portion of the virtual address space, instead
of sharing the same pages in the physical address space.  The way to
get page level copy-on-write in the 386 is:

	1) Map the new segment to another, unused portion of the virtual 
	   address space.

	2) Copy the page table entries from the old segment's part of the
	   page table into the new segment's part of the page table, and
	   mark all the new entries read-only.

	3) Whenever you get a fault, copy the old physical page into a
	   new one, and replace the page table entry with one for the
	   new page, and mark the page as writeable.

   This is less aesthetically pleasing than per-segment page tables, since 
you have to poke around a bit whenever you get a fault, to figure out
whose segment it's in, and why you got the fault.  It IS doable, though,
and I stand corrected.

   One more thing:  I don't think it's very appropriate to call people
stupid when they are wrong, as long as they are being reasonably polite
about it.


-- 
    Dan Frank
    uucp: ... uwvax!prairie!dan
    arpa: dan%caseus@spool.wisc.edu