From owner-freebsd-hackers  Tue Mar 28 16:56:47 1995
Return-Path: hackers-owner
Received: (from majordom@localhost) by freefall.cdrom.com (8.6.10/8.6.6) id QAA01611 for hackers-outgoing; Tue, 28 Mar 1995 16:56:47 -0800
Received: from vitruvius.arbld.unimelb.EDU.AU (vitruvius.arbld.unimelb.EDU.AU [128.250.136.1]) by freefall.cdrom.com (8.6.10/8.6.6) with ESMTP id QAA01604 for <freebsd-hackers@freefall.cdrom.com>; Tue, 28 Mar 1995 16:56:44 -0800
Received: (darrenr@localhost) by vitruvius.arbld.unimelb.EDU.AU (8.6.9/8.6.4) id KAA00923; Wed, 29 Mar 1995 10:55:19 +1000
From: Darren Reed <darrenr@vitruvius.arbld.unimelb.edu.au>
Message-Id: <199503290055.KAA00923@vitruvius.arbld.unimelb.EDU.AU>
Subject: Re: Configuring driver added via LKM
To: dufault@hda.com (Peter Dufault)
Date: Wed, 29 Mar 1995 10:55:17 +1000 (EST)
Cc: dufault@hda.com, freebsd-hackers@freefall.cdrom.com
In-Reply-To: <199503290024.TAA00578@hda.com> from "Peter Dufault" at Mar 28, 95 07:24:18 pm
X-Mailer: ELM [version 2.4 PL23]
Content-Type: text
Content-Length: 1056      
Sender: hackers-owner@FreeBSD.org
Precedence: bulk

In some email I received from Peter Dufault, they wrote:
> 
> Is it on purpose that "id_alive" isn't set if you are reconfiguring
> a device in "isa/isa.c"?  Is there an entry point for "I'm about
> to remove you"?

> How completely thrashed out is registering a device with the kernel
> (completely, including using Garrett's devconf routines) and then
> de-registering it in order to completely clean up and register
> again?

> I may be missing something, but I don't see a set of entry points
> that look like there are many people adding and removing device
> drivers for debug.  I'd like the device driver code to remain
> completely clean without any conditionals for the LKM.
> 
> If anyone has any pointers I'd appreciate them.

You might want to do the unload of the LKM in stages...first, shut the
device down from further IO requests (open/read/write), allow outstanding
ones to complete and then remove it from the various tables.  I can't
see why this wouldn't be doable, as long as you weren't expecting the
unload to be immeadiate.

darren