Christopher R. Bowman wrote:
> 
> On Thu, 31 Jul 1997, Eric C Wagner wrote:
> 
> > 
> > I just got 1.2.1 working on a IIci with cache. The problem I'm having is

[problems most likely due to the ncrscsi driver deleted...]

> My memory is vague, but there is a kernel option that deals with turning
> the cache on or off, are you using a custom kernel or a generic.  Whoops

He's probably using the 1.2.1 distribution kernel.  There is an option
concerning the IIci cache.  More on this below.

> back up there, you are using fsck to manually fix the disk?  As I recall
> and I am sure that some one will correct me if I am wrong, fsck
> uses the raw disk, bypassing the kernel buffer, if some of the bogus
> blocks that are getting fixed by fsck are already in the buffer cache,
> then when they get flush so will the changes fsck made to fix your disk.
> As I recall you have to use shutdown -n to shutdown with out flushing
> the disk buffer cache right after you twidle the disk with fsck.  Can
> someone else sanity check me that this is right.

Quite true.

> I think the IIci cache is working ok, my IIci is running right now so
> I can't pry it open and check for you, but I can't find the card
> in my drawer where I would have stored it if I wasn't using it.

Unless you are using a custom kernel, the IIci cache isn't working at all.
The default for GENERIC kernels is to have it turned off.

I've done some experimenting with the external cache on the Daystar
accelerator for my IIci, and the kernels which actually boot seem to be a
bit more crash prone than usual.  Well, they often hang, occasionally with
"value modified on freelist:  0xdeadbeef does not equal 0xdeedbee8" or
something similar.  For some reason, it doesn't seem to like the 1 line
writeback cache on the L2 cache.  If I turn that off, it _seems_ to work.
I say seems because I can't actually tell any difference in performance
:-( and dhrystone can't either.

Oh well, I guess I'd better stop rambling.

Later.

-- 
Colin Wood                                 cwood@ichips.intel.com
Component Design Engineer - MD6                 Intel Corporation
-----------------------------------------------------------------
I speak only for myself, not for my employer.