Dmitry Torokhov wrote:

> You need to let go of the model that driver that drives hardware also
> do the device discovery and it will all fall into place.

It's not possible to sanely let go of that model. ISA discovery involves 
(as an example) poking at the same I/O ports as using does, meaning you 
will have to share a request_region() over discovery and use for one -- 
you can't decouple that due to obvious races.

>> Anyways, the additional method would, I feel, be the conceptually
>> cleanest approach. Practically speaking though, simply doing a manual
>> probe and only calling platform_register() after everything is found to
>> be present and accounted for is not much of a problem either.
>>
> 
> Unfortunately it breaks manual driver binding/unbinding through sysfs
> so I don't think it is a good long-term solution.

Yes. I don't see a significantly cleaner solution then than the slightly 
hackish "using drvdata as a private success flag" that I posted before. 
Example patch versus snd_adlib attached again. This seems to work well.

Takashi: do you agree? If the probe() method return is not going to be 
propagated up, there are few other options.

(Continuing the loop on IS_ERR(device) is then also a bit questionable 
again as the IS_ERR then signifies an eror in the bowels of the device 
model, but I feel it's still the correct thing to do)

Rene.