Re: 2.5.69-mm5: reverting i8259-shutdown.patch

[Pavel Machek <pavel@ucw.cz>]

Hi!

> > > Hi again, Andrew,
> > > 
> > > Besides the "make_KOBJ_NAME-match_BUS_ID_SIZE.patch" causing "pccard"
> > > oopses, I've also found that, with 2.5.69-mm5 compiled with ACPI
> > > support, my laptop is unable to power off. The kernel invokes
> > > "acpi_power_off" and stays there forever.
> > > 
> > > I've found that reverting the "i8259-shutdown.patch" fixes the problem
> > > and my laptop is able to shutdown properly (init 0) when using ACPI.
> > > 
> > > A hardware bug? A kernel bug?
> > 
> > And thanks again, again.
> > 
> > That's the below patch.  It looks pretty innocuous.  I'd be assuming that
> > there's something in the shutdown sequence which needs 8259 functionality
> > after the thing has been turned off.
> > 
> > This could well depend upon .config contents and linkage order.
> > 
> > Eric, maybe we need to turn it off by hand at the right time rather than
> > relying on driver model shutdown ordering?
> 
> Interesting. This is yet more proof that system-level devices cannot be
> treated as common, everyday devices. Sure, it's nice to see them show up
> in sysfs with little overhead, and very nice not to have to work about
> them during shutdown or power transitions. But there are just too many
> special cases (like getting the ordering right ;) that you have to worry
> about.
> 
> So, what do we do with them? 

I guess shutdown needs to be treated like suspend, and needs to have
"level". There should be no shutdown, you should do suspend(5, ) and
go through all levels properly.
								Pavel

-- 
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]