[Pkg-acpi-devel] acpid module loading

[Loïc Minier]

On Thu, Oct 30, 2008, Michael Meskes wrote:
> On Wed, Oct 29, 2008 at 05:34:27PM +0100, Loïc Minier wrote:
> >  He suggested improved the wording in the default file, and I noticed
> >  additional issues in the init script to enhance / fix it further.
> 
> Do your changes fix 502704? Or do you think they might fix it, I take it you
> don't have any feedback from Derrick either.

 They could, because I don't understand the bug, but I'm not sure they
 do.

 Julien reported set -x during boot to end with the "modprobe --all"
 failing; I'm not sure the method of looking at the set -x output during
 boot was reliable enough and don't understand why the call would fail
 except for an empty list of modules with some whitespaces, but we
 confirmed his list was empty.

> >  - The list of modules which are listed by default in MODULES is
> >    probably obsolete: these modules exist, but what about new modules,
> >    or modules which might be added in the future?  e.g. bay, dock, sbs
> >    etc.
> They might have to be added.

 This is inherently fragile because the same acpid could be used with
 multiple kernels.   :-/

> >  - This is all more shell which gets run before acpid is started.
> > 
> >  As a result, I'd like to propose dropping module loading entirely from
> >  acpid.init and relying on udev to load modules based on what's present
> >  on the system and not supported by the kernel (current behavior anyway:
> >  the modules are already loaded when the acpid init script starts!).
> 
> Haven't tried but sounds logical to me. Big question is do we really want to
> make an update like this for Lenny? If not we should wait until we have fixed
> 502704 IMO.

 I'm pretty sure it would either fix 502704, or would make it clear that
 acpid has a more serious bug if it's started and exits because
 kernel/module are not yet setup.

 I agree the slope for the lenny updates has been slippery with the init
 changes.

> >  Another open point is why do people get acpid start errors on boot and
> >  not on invoke-rc.d acpid start after boot.  No clue here.  One guess
> >  could be that acpid starts too early, but this seems weird to me.  If
> >  that's the case though, I think we should patch acpid to wait for stuff
> >  to appear instead of failing.  I've requested more info to people
> >  seeing this for now.
> So this is #502704 I guess. Or do you have more reports like this?

 Julien's has the same bug; he is using a seft built kernel with ACPI
 stuff builtin.

-- 
Loïc Minier