*BSD News Article 18373


Return to BSD News archive

Newsgroups: comp.os.386bsd.development
Path: sserve!newshost.anu.edu.au!munnari.oz.au!constellation!osuunx.ucc.okstate.edu!moe.ksu.ksu.edu!ux1.cso.uiuc.edu!newsrelay.iastate.edu!news.iastate.edu!ponderous.cc.iastate.edu!michaelv
From: michaelv@iastate.edu (Michael L. VanLoon)
Subject: Re: boot.c for kernel > 1M
Message-ID: <michaelv.742631490@ponderous.cc.iastate.edu>
Keywords: Mach, NetBSD
Sender: news@news.iastate.edu (USENET News System)
Organization: Iowa State University, Ames IA
References: <27081@hq.hq.af.mil>
Date: Wed, 14 Jul 1993 06:31:30 GMT
Lines: 39

In <27081@hq.hq.af.mil> john@Pt.hq.af.mil (John Ryan) writes:

>We are trying to merge NetBSD  and Mach boot sequences. I wanted
>some advice from people who might be trying to patch NetBSD to
>load it above 1M.

>Question 2) Has anyone calculated the text, bss, data, and symbol offsets
>	for > 1M?  That is Are you going to want things a specific
>	locations.

I would think it would be best to have it as low as possible, hence,
right at the 1M boundary.  My reason for this is that I sometimes
throw in a ISA bus extended memory card to boost the memory on my
system.  This card is extremely slow, but generally faster than
swapping.  I would like to be reasonably well assured that my kernel
is never going to get stuck in slow memory, even if some applications
do.  This is, I think, a reasonable request.

What may be more difficult, but nevertheless desirable, is to have the
boot code determine at startup time different areas of fast and slower
memory.  Anyone who's ever used QEMM seriously has seen this in
action.  It would allow the kernel to weight its decision about where
to swap things in by how fast the memory location is.  Or,
alternatively, to stick disk buffers or something similar in slower
memory, since even slower memory is generally faster than disks or
tapes.

Alternatively, you could maybe tell the system how your memory is
layed out in your config file and build it into your kernel as a
static table, which would relieve it of the task of gauging the speed
of your memory at boot-up.  But would still allow the VM mechanism to
weight paging locations by memory speed.  Or, maybe would allow the
physical I/O buffers to be stuck in slower memory.

-- 
------------------------------------------------------------------------------
  Michael L. VanLoon                           Project Vincent Systems Staff
  michaelv@iastate.edu              Iowa State University Computation Center
------------------------------------------------------------------------------