Hi Jon,
On 27 May 2014 09:42, Jon Loeliger loeliger@gmail.com wrote:
The preferred device can be specified with a DM_FLAG_PREFER flag or a 'dm,prefer' property in the device tree node.
It is possible that a better approach will come to light in the future, but this gets around the problem as it currently stands.
Here's your clue that something isn't quite right here. Again, this looks like the us of a DTS property to describe a SW functionality rather than describe the HW itself.
I'm not sure what needs to be done here either, but maybe it centers on a better understanding of *why* something is being preferred? And what if it is "preferred", then what does that really mean? And what if two devices are labelled as "preferred" -- Is that OK? Who checks and enforces it?
Sure, this may be a hack for now, but it needs more thought.
Agreed, but I've done all the thinking I can so far, and this is the result.
Another way of doing this would be to designate the platform data device as a 'backup' device, only to be used if we don't get a similar device from the device tree. But I'm not sure that it would be better that way.
This is not intended to be a widely-used mechanism - it just deals with the issue of device tree vs. not.
My preference would be to require that all devices be represented by a device tree with driver model. However, I think that would discourage adoption in the short term. Also for sandbox we really do want to test all the code, and so always having a device tree would reduce test coverage.
Regards, Simon