White/blacklisting manf_id/prod_id vs. func_id

[Russell King]

On Tue, Nov 21, 2006 at 08:25:21AM -0500, Dominik Brodowski wrote:
> On Tue, Nov 21, 2006 at 08:59:11AM +0000, Russell King wrote:
> > The problem seems to be that even for _good_ devices, we're still
> > having to add their IDs to the tables.  That's absolutely ludicrous.
> 
> That's not true.

That does seem to be what is happening however.

> > Good devices should just work out of the box as they used to with
> > the original pcmcia-cs implementation.
> 
> They do work out of the box. They just might not work if userspace is
> not up and running -- but that's not a regression. So please stop
> making this claim.

Why then, is this Elan card not detected out of the box despite advertising
that it is a serial function, and the serial_cs driver advertising that
it takes care of devices which advertise that they're a serial function.
But when the explicit IDs are added, it works.  _That_ sounds to me very
much like the function matching is broken.

> > What are end users supposed to do when they buy new cards?  Build
> > their own kernels?  Scream at their vendor to merge a patch which
> > adds their card IDs to a driver?
> 
> What's the story with PCI drivers and devices?

99% of them just work.  We occasionally have to add another ID, and
recently there have been some specific IDs we've been able to remove.
On the whole, the list of explicit devices is actually rather stable.

> One thing which might be worth thinking about is adding an interface
> similar to PCI's new_id. Patches are welcome.

Don't look at me - remember I gave up PCMCIA and recently serial (and
MMC) as well.  But that doesn't mean I give up reviewing serial patches,
which is what this is all about.

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