Subject: Re: time for INSTALL.ACPI ?
To: Manuel Bouyer <bouyer@antioche.eu.org>
From: Steven M. Bellovin <smb@cs.columbia.edu>
List: port-i386
Date: 11/22/2006 09:22:44
On Wed, 22 Nov 2006 12:34:40 +0100
Manuel Bouyer <bouyer@antioche.eu.org> wrote:

> On Mon, Nov 13, 2006 at 09:41:35PM +0100, Manuel Bouyer wrote:
> > On Mon, Nov 13, 2006 at 09:15:49PM +0100, Pavel Cahyna wrote:
> > > On Mon, Nov 13, 2006 at 08:46:57PM +0100, Manuel Bouyer wrote:
> > > > On Mon, Nov 13, 2006 at 07:09:29PM +0000, Patrick Welche wrote:
> > > > > Would it be safe to simply switch on acpi in INSTALL?
> > > > 
> > > > The main problem is that there are some brocken ACPI
> > > > implementations out there (and still some bugs left in our ACPI
> > > > code I guess). So we really need both ...
> > > 
> > > Would "disable acpi" in userconf work?
> > 
> > Hum, maybe. I'll investigate that.
> 
> OK; I've built a release with the attached diff and tested on 3
> different systems:
> - recent P4 HT: no problems using ACPI+ioapic, and both can be
> disabled with boot -c; we get the same dmesg as INSTALL kernel
> without the patch
> - older P2 system: the kernel finds ACPI but no IOAPIC; the device
> uses legacy interrupts and are functionnal.
> - a system without ACPI (a Xen HVM guest actually): the kernel prints
> some messages about not being able to find ACPI and them boots in
> legacy mode. No problems either.
> 
> So it looks like adding ACPI+ioapic to INSTALL is workable: on recent
> systems is should work, on non-acpi one it works; and on sytems where
> ACPI is broken is should be possible to disable it with boot -c.
> The only problem is that with my patch bootfloppy-big overflows
> again :(
> 
> Comments ? Any objection to this change (once a solution for
> bootfloppy-big is in place) ?
> 
Have you tested on a laptop with APM?  I ask because I have to put

	no acpi0                at mainbus0

in my (T42) laptop config file (which is mostly an include of
GENERIC_LAPTOP).  However, I don't recall the failure mode if I don't
have it -- it might just be that APM doesn't work, which doesn't matter
for INSTALL kernels.


		--Steve Bellovin, http://www.cs.columbia.edu/~smb