Diversity mode on 8385

Dan Williams dcbw at redhat.com
Tue Mar 11 08:58:04 EDT 2008

On Tue, 2008-03-11 at 08:38 +0100, Holger Schurig wrote:
> > No, but until we either fix WEXT or add this stuff to
> > cfg80211, we shouldn't be adding stuff to iwpriv, because
> > every driver that uses iwpriv does the shit differently, and
> > that just sucks.
> Ah, iwpriv worked for the last 10 years.

For command line static setups, yes.  For everyone else trying to do
anything, no.

> Also suppose every driver would now use debugfs for diversity. 

The point of doing it in debugfs is that there is _no_ guarantee of API.
And the users that are using these iwpriv options are usually able to
control their own environment (ie testing, embedded, academic, etc) and
therefore should be quite able to understand how to use debugfs.

> Then I'm sure that every driver would implement this 
> differently, too. The differences of driver-implementation has 
> nothing to do with the mechanism of the user-space-API, it has 
> something to do that most driver developers only really care for 
> their device and fear modifying other devices (to unify the 
> API), because often they don't have those other devices or don't 
> understand their complexity.

I simply refuse to ack patches that add things to iwpriv APIs upstream.
That's just not going to happen.  Either keep it in your private git
repo, or fix cfg80211.  Do not put anything into iwpriv.

> > So obviously we need to fix WEXT; but if you don't want to
> > wait until then, I'd suggest debugfs.
> Again, debugfs is for debugging. That's why it is called debugfs 
> and not wextreplacefs. Embedded devices have size constraints 
> and usually don't compile it in.
> You can use sysfs or module-parameters for such things.

You should not be using sysfs, because sysfs is kernel API and usually
cannot change.

You probably don't want to use module parameters, because you don't get
change events when the user changes the value.  If that isn't an issue,
then a module parameter is usually OK.

> But anything that might be needed for regular operating of a 
> device and which depends on debugfs should get a NACK. At least 
> it get's one from me --- there are alternatives, we should use 
> them.

Well, I'm going to NAK iwpriv stuff...  If there's a deficiency in
cfg80211 or WEXT, we should fix that inconsistency.  We do not add
iwpriv commands.


More information about the libertas-dev mailing list