From owner-freebsd-hackers  Wed Jun  7 17:28:42 1995
Return-Path: hackers-owner
Received: (from majordom@localhost)
          by freefall.cdrom.com (8.6.10/8.6.6) id RAA24525
          for hackers-outgoing; Wed, 7 Jun 1995 17:28:42 -0700
Received: from cs.weber.edu (cs.weber.edu [137.190.16.16])
          by freefall.cdrom.com (8.6.10/8.6.6) with SMTP id RAA24497
          ; Wed, 7 Jun 1995 17:27:12 -0700
Received: by cs.weber.edu (4.1/SMI-4.1.1)
	id AA07812; Wed, 7 Jun 95 18:19:01 MDT
From: terry@cs.weber.edu (Terry Lambert)
Message-Id: <9506080019.AA07812@cs.weber.edu>
Subject: Re: 2.0.5-A: Very disheartening?
To: phk@ref.tfs.com (Poul-Henning Kamp)
Date: Wed, 7 Jun 95 18:19:01 MDT
Cc: uhclem%nemesis@fw.ast.com, jgreco@brasil.moneng.mei.com,
        hackers@FreeBSD.org, bugs@FreeBSD.org
In-Reply-To: <199506072203.PAA03872@ref.tfs.com> from "Poul-Henning Kamp" at Jun 7, 95 03:03:30 pm
X-Mailer: ELM [version 2.4dev PL52]
Sender: hackers-owner@FreeBSD.org
Precedence: bulk

> > > As far as I can see, the only failing part is the mechanisms associated
> > > with a compressed kernel, or the compressed kernel itself and certain
> > > types of cache subsystems.   I can boot from a floppy with a uncompressed
> > > kernel and that works fine too.  Only the compressed kernel has trouble.
> > 
> > This makes sense, actually.
> 
> Explain please...

Split I and D would cause this.

Also, second reference cached code (uncompressed would be first
reference, so the cache would be clean) on systems with broken caches.

Even on working caches, you'd probably want to either set the pages you
are loading the compressed image into non-cacheable, or BINVD the target
pages from the compression to insure that you don't have bad code in your
cache.

The instruction prefetch crap, and the undocumented-but-existing cache
crap is described in detail in "The Undocumented PC" (one of my favorite
PC computer reference works).

More general code that world detect cache writeback failure and other
issues would be a good thing, but would probably mean two-staging the
boot to make it happy.  It could certainly be useful for HiNT chipset
NiCE EISA motherboards and other boards that lie about cache writeback.

The instruction prefetch cache issue should be an easy one to test for...
are any of the following true:

o	All failures of this type have been on Pentium boxes
o	All failures of this type have been on non-Pentium boxes

If the latter, it's instruction prefetch cache flushing that's needed.
If the former, it's another issue.


					Terry Lambert
					terry@cs.weber.edu
---
Any opinions in this posting are my own and not those of my present
or previous employers.