linux 2.6 pcmcia power issues

[Russell King]

On Sun, Feb 27, 2005 at 01:45:10PM -0800, David Hinds wrote:
> On Sun, Feb 27, 2005 at 05:06:22PM +0000, Russell King wrote:
> > On Sun, Feb 27, 2005 at 06:03:34PM +0100, Elmar Athmer wrote:
> > > Hi
> > > | That's because 2.4 doesn't report the problem.
> > > 
> > > But 'cardctl info' gives useful information with 2.4, 2.6 reports only
> > > empty values.
> > 
> > I repeat.  2.4 kernels do not warn when they are unable to remove
> > socket power.
> > 
> > cardctl <anything> also won't tell you when this condition exists.
> 
> Has anyone ever observed a situation where the "unable to remove
> socket power" message did not indicate that the entire CardBus bridge
> register set was inaccessible due to a bad memory mapping?

Take a look at pcmcia.arm.linux.org.uk - I kept a set of links to
the original reports in the archives on there.  There do appear to
be a couple of instances there where there is no obvious overlapping
mapping between the CardBus bridge registers and system memory.

I suspect that the Linux kernel is missing a great load of resource
information which Windows may have - Dominik has hinted that there's
ACPI tables which give the locations of magic IO/memory reservations
which no Linux kernel takes account of.

I also suspect that it's not worth trying to investigate these reports
much further than the "does the cardbus bridge regions overlap real
memory" case until such time that the Linux kernel does take account
of those reserved regions.

(This also reminds me to update my signature...)

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core