On Mon, 31 Aug 2015, Stephen Rothwell wrote:

On Mon, 31 Aug 2015 17:48:42 +1000 Stephen Rothwell [off-list ref] wrote:

quoted

On Mon, 31 Aug 2015 09:04:22 +0200 Ricard Wanderlof [off-list ref] wrote:

quoted

On Fri, 28 Aug 2015, Mark Brown wrote:

quoted

On Fri, Aug 28, 2015 at 09:40:41AM +0200, Ricard Wanderlof wrote:

quoted

On Fri, 28 Aug 2015, Stephen Rothwell wrote:

quoted

In fact the exact same construct is used by a handful of other codec 
drivers which apparently don't fail.

quoted

I'm suspecting something slightly more convoluted like a missing 

#include .

No, the issue is that you have used a different variable name when 
declaring the IDs and when referencing them in the module device table.

Yeah, I realized that upon closer inspection. 

What bugs me is that my ARM gcc didn't seem to flag this, whereas the 
x86 gcc did upon subsequent testing. And yes, CONFIG_OF is set during my 
build.

Do you have CONFIG_MODULE set in your build? (just guessing)

Actually what matters is if you build the driver as a module or not.
See include/linux/module.h and the definitions of MODULE_DEVICE_TABLE().

Bingo.

Haven't verified that, but it's true, the kernel build for our ARM system 
is largely monolithic as we have no need to reconfigure it once it has 
been built. Whereas in my x86 test build the driver was built as a module.

Thanks Stegphen!

/Ricard
-- 
Ricard Wolf Wanderlöf                           ricardw(at)axis.com
Axis Communications AB, Lund, Sweden            www.axis.com
Phone +46 46 272 2016                           Fax +46 46 13 61 30