*BSD News Article 12926


Return to BSD News archive

Path: sserve!newshost.anu.edu.au!munnari.oz.au!news.Hawaii.Edu!ames!network.ucsd.edu!romulan!rdante
From: rdante@sdcc13.ucsd.edu (Rick Dante)
Newsgroups: comp.os.386bsd.bugs
Subject: Re: This is a strange one!
Date: 17 Mar 1993 05:19:00 GMT
Organization: Newsreaders Anonymous
Lines: 41
Distribution: world
Message-ID: <1o6cc4INNjuf@network.ucsd.edu>
References: <C40400.9qv@veda.is>
Reply-To: rdante@sdcc13.ucsd.edu
NNTP-Posting-Host: dialin1-47-1.extern.ucsd.edu

In article 9qv@veda.is, root@veda.is (Tree and Root) writes:
> lessen@axion.bt.co.uk (Lee Essen) writes:
> 
> >whenever I try to use any of those binaries (disklabel, dd or cat) with
> >any parameters (not even accessing the drive) the core dumped.
> 
> >I have to rebuild them to fix the problem!
> 
> >Weird! - Anyone got an explaination for that!
> Here's one possible explanation:
> 
> The binaries are corrupted because other random fileblocks have replaced
> parts of the binary file. This is reported to be due to the FS buffer cache
> being inconsistent with the FS. It usually has no lasting effects and just
> causes random failures, but unfortunately files do sometimes sustain real
> and permanent damage. Temporary (but highly visible) damage is widespread
> had to reboot or type 'make' several times (once after each compilation
> error) or the machine just hung or crashed during the compile of libc.a.
> 
> Do any solutions exist yet, or has the offending code at least been located?
> Adam David (adam@veda.is)

(much text deleted)

Have you tried disabling your external cache? At the suggestion of a previous article I 
disabled my secondary cache and I am able to do a make in /usr/src straight through with 
no problems (except that it's 2.5 times slower :( ). Many people seem to have solved 
their problems by replacing cache chips. I did this 3 times without luck so I think it 
might be the cache controller but I'm not sure. I now have a good excuse to replace my 
motherboard (and get a 486).  To test out the FS buffer cache corruption theory with my 
external cache enabled after getting compilation errors I would go into /tmp (which I 
have a memory filesystem mounted on) and copy big files until I hear disk swaping (which 
suggests that the FS buffer cache is being shrunken to make way for files in the MFS). I 
then start make again and no problems (where otherwise I would have to reboot). All this 
does is help confirm the FS buffer cache incoherency problem but I feel this problem is 
caused by the external cache of some machines and that alone. Other opinions might differ however.

Rick Dante