Re: OMAP34xx

[Tony Lindgren <tony@atomide.com>]

* Russell King - ARM Linux <linux@arm.linux.org.uk> [120205 10:10]:
> On Sun, Feb 05, 2012 at 10:33:16AM -0800, Tony Lindgren wrote:
> > * Russell King - ARM Linux <linux@arm.linux.org.uk> [120205 09:37]:
> > > So in the meantime, people should put up with the kernel reporting an
> > > "Error" at error-message level at boot time because they didn't configure
> > > something?
> > > 
> > > No, it needs fixing, because it doesn't justify being an error.  It's
> > > wrong, plain and simple.  Again, if you don't want to send it during
> > > -rc, I'll send it to Linus as a patch for him to decide whether he wants
> > > to take it as -rc material.
> > 
> > Hmm, maybe I misunderstood you.
> > 
> > Certainly fixing the "Error" makes sense for -rc, but are also thinking
> > about adding error checking to all platform_device_register() calls?
> 
> I do think they're valid for -rc, because should it fail, things won't
> work as one desires.

OK, this easily gets into the area where we might get flames from
Linus like "fixing features during -rc that never worked properly
before"..
 
> I did stop short of fixing it properly - and by "properly" I mean that
> if we fail to register the platform data for the device, then we should
> not register the device itself.  That involves a change of behaviour in
> the setup of the system which _might_ cause a regression.
> 
> Whereas, merely printing an error for a failure to register a device
> wouldn't be a regression (it might uncover another error, but that's
> arguably a good thing.)

OK that sounds safe to me.
 
> The last bit of "properly" which needs discussion is concerning whether
> we should even be trying to register this devices platform data and
> device structure if WL12XX_PLATFORM_DATA is not enabled - if this is
> not enabled the platform data won't be stored, and presumably the
> wireless driver will be missing some vital platform data.

That sounds like the way to go to me.
 
> It seems silly to have the device registered without the platform data,
> so that if the module is built for the kernel, it could be inserted and
> bound to that device...
> 
> So even for this apparantly simple looking issue, there's some searching
> questions that need answering.

I doubt that there's anything else behind it except trying to leave
out some ifdefs.

Regards,

Tony