PCMCIA Bluetooth Card (AnyCom CF - 300) on ARM

[Dominik Brodowski]

Hi Russell,

On Wed, Sep 28, 2005 at 05:05:35PM +0100, Russell King wrote:
> On Tue, Sep 27, 2005 at 06:55:12PM +0200, Dominik Brodowski wrote:
> > On Tue, Sep 27, 2005 at 07:36:58PM +0300, Itzhak Ben-Akiva wrote:
> > > cardmgr[841]: error in file './config.opts' line 20: invalid port range
> > > 0x60000000-0x60ffffff
> > > cardmgr[841]: no pcmcia driver in /proc/devices
> > > 
> > > include port 0x600-0x60f, though, went ok.
> > 
> > That's a cardmgr bug. Then use this small sysfs shell script instead
> 
> Actually, it's worse than that.
> 
> Sep 28 16:48:43 cps cardmgr[2350]: watching 2 sockets
> Sep 28 16:48:43 cps cardmgr[2350]: could not get CS revision info!
> Sep 28 16:48:43 cps cardmgr[2350]: exiting

Old cardmgr version (3.1.22). With fairly recent pcmcia-cs versions,
it still works.

> With 2.6.14-rc2, cardmgr can't even start up, so it can't register
> any memory or IO regions.  Effectively, this has broken the ioctl
> interface _right_ _now_.

No. Not if you use somewhat recent pcmcia-cs.

> I also believe you are raising the bar to having working PCMCIA too
> much by deprecating the ioctl method in such a short space of time.
> Embedded folks take _years_ to adopt new ideas - some of them haven't
> even adopted things like udev or hotplug yet.

Well, PCMCIA is definitely something many people blame for being broken,
often even if it is unjustified and e.g. the PCI layer broke. Therefore,
with PCMCIA integration into the kernel still being quite behind, I do not
see a chance for slowing down _just_ _now_.

> You can't force them to do so either - they'll stick with some random
> old kernel version in preference to upgrading to something new.

Then let's start 2.7. Supporting both cardmgr and in-kernel matching is
quite difficult. Only if cardmgr support is dropped, for example, we can
finally allow "rmmod"ing drivers which are currently bound to devices. If
2.7. doesn't start soon, though, I intend to disallow CONFIG_PCMCIA_IOCTL
in Kconfig for 2.6.15 (but do not remove actual support for it), and remove
it for 2.6.16.

> Can pcmciautils at least work without udev?

pcmciautils can work with _either_ udev or hotplug.

In addition, if you
a) build all stuff into the kernel
b) the socket driver doesn't need a resource database,
c) all devices used have PROD_ID/MANF_ID/CARD_ID matches, and
d) require no CIS update,
you do not need pcmciautils or pcmcia-cs at all. In fact, with some hacking
you can get rid of c) if you know what types of cards get inserted (e.g.
only CF IDE cards). Therefore, most embedded stuff should work out of the
box, AFAICS.

	Dominik