*BSD News Article 53793


Return to BSD News archive

Newsgroups: comp.unix.bsd.freebsd.misc
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!simtel!oleane!tank.news.pipex.net!pipex!news.mathworks.com!newsfeed.internetmci.com!EU.net!sun4nl!news.iaf.nl!news.es.iaf.nl!yedi!wilko
From: wilko@yedi.iaf.nl (Wilko Bulte)
Subject: Re: Parity error...
Organization: Private FreeBSD site - Arnhem, The Netherlands - The Deamon is Free! -
Message-ID: <DGywC9.oH@yedi.iaf.nl>
References: <46i53h$5k0@news.bu.edu>
Date: Tue, 24 Oct 1995 19:19:21 GMT
Lines: 38

mi@bu.edu (Mikhail Teterin) writes:

>Happened to work on a new SGI for the last 2 weeks running IRIX 5.3
>It happened ones, that machine suddenly beeped, and put on a window,
>saying that it noticed a parity error. It even gave a SIMM number.
>I recalled last time it hapened on my FreeBSD box (panic, etc.), and
>prepared to reboot. To my surpise nothing happened after I pressed Ok.
>I don't even think any process was killed... It just continued to work.

>Is this a hard to implement, very much hardware related feature?

NOTE: this might be crystal ball gazing, lacking any hard data
on the SGI:

SGI could have implemented a register in the machines hardware that
'traps' the address of the memory location that caused the parity
error. Using this register the kernel could do a physical -> virtual
translation to see which memory page this corresponds to. 

When this page was not in use, one could mark the page unavailable
to the system and continue happily. Try this again when the error
was in the kernel address range...

Alas, PCs have lame enough hardware that only pulls on the NMI
line in case of a parity error. No register. I've seen kernel 
source that tries to find the error by doing a complete sweep
of memory and see if the error re-occurs. Experience showed 
only +- 20% of the errors were 'hard' enough to be found this way.

For the record: the PDP11/34 (of 1976 vintage) had such a register..
Modern PC mainboards don't even use parity anymore (memory technology
is assumed to be flawless (?))

Bottom line: there is little chance FreeBSD could ever be like the
SGI

Wilko