[pkg-wpa-devel] Driver autodetection

[Kel Modderman]

Reinhard Tartler wrote:
> On Wed, Apr 05, 2006 at 07:40:18PM +1000, Kel Modderman wrote:
>   
>> Defaulting to "wext" will prove to be more and more true as time goes 
>> on. This will be true for almost all in-kernel modules in the future.
>>     
>
> True, this is the long to midterm solution.
>
>   
>> How can we pretend to guess a driver for a interface created by an 
>> external-to-kernel module, across many different user configurations 
>> (driver version, kernel version etc) with any sort of precision? and 
>> also predict what wpa_supplicant driver backend they can use in the 
>> future, in the case that a user updates to a newer version of some 
>> driver? I'd think this to be a very fragile process until I see some 
>> tried and tested code.
>>     
>
> I agree to you that we cannot support $RANDOM user compiled kernel
> drivers, but as package maintainers, we work on a distribution. The
> distribution provides some drivers, so I think this bug is about the
> supporting available drivers we currently have.
>   

A snapshot of a system could work well with the proposed detection. But 
things would change when modules are upgraded, kernel is upgraded etc.
> Hardware detection and activation currently works quite nice on modern
> kernels with udev. My proposal is to exploit the infrastructure we
> currently already have, and add some hooks to give the wpasupplicant
> package hints which driver backend to use.
>   

Sure, I understand the aim, but just disagree with this approach, to an 
extent.
>   
>> My opinion is; its not too hard to add wpa-driver foo to the 
>> configuration, so long as documentation is good.
>>     
>
> This is the fall back. This bug is because it should not be necessary to
> specify wpa-driver. It should work out of the box.
>
> Given that I don't see much need for that, I don't give this bug much of
> a priority. I rather outlined how I could image a solution for this, so
> that we can finally close this bug.
>   

Maybe one day it the driver backend could be determined at some low 
level, but that should be done upstream, imho. I just see too many 
difficulties with "guessing" the answer during this turbulent time for 
wifi on linux.

Thanks, Kel.