Path: utzoo!utgpu!utstat!jarvis.csri.toronto.edu!mailrus!cornell!rochester!udel!mmdf
From: Leisner.Henr@xerox.com (marty)
Newsgroups: comp.os.minix
Subject: Re: PC Minix/64KB limit
Message-ID: <8832@louie.udel.EDU>
Date: 15 Feb 89 19:17:26 GMT
Sender: mmdf@udel.EDU
Lines: 54


Once again, the Minix-PC memory models raises it's ugly face.

Without even mentioning the effort required to write a more complicated
80x86 C compiler which can support near and far data types, all of Minix
runs on the implicit assumption all message data is in the data segment.  I
don't like it, but that's the way it is.

Without doing much modification to the a.out header structure, I wanted to
define a large data model which would allow:
	<= 64 K text
	<= 64 K data
	<= 64 K stack
	infinite (well, almost) number of 64K additional segments.

Assuming we have some way of memory management subroutines and a new,
improved malloc for this large data model (something similar to the way
286/unix works)
the operating system has to run with 32 bit virtual addresses -- there is
an implicit assumption in Minix that messages are addressed in the data
segment and 16 bit pointers are used.

Would the above architecture allow more complicated programs?  It seems
with a decent C comiler, 64k of text is enough for a lot.  Also, the above
architecture would allow support for:
	1) shared memory segments
	2) embedding strings in the text segment

On protected mode 286 machines, the stack could also grow dynamically up to
64k. 
In addition, protected mode implementation can give individual processes
huge amounts of heap space (provided there is enough ram) without resorting
to swapping. 

In addition, the memory manager/kernel interface has to be rearchitected,
along with how memory manager information is encoded (right now we have
NR_SEGS in const.h).

I'm not sure the changes necessary to fully support the 8086 family is
worth it.  On a 32 bit 68xxx or 32bit 386 version, all these problems go
away.  I suppose due to the size of the market, the Ms/Dos compiler vendors
have a lot of incentive to make the Intel architecture do more.


Does anyone have a copy of PC-Minix which uses 32bit pointers within the
operating system?  I looked at it and felt it would cause distributed
changes to the whole system -- I never got around to it.  


marty
ARPA:	leisner.henr@xerox.com
GV:  leisner.henr
NS:  martin leisner:wbst139:xerox
UUCP:  hplabs!arisia!leisner