On Mon, Oct 20, 2014 at 08:16:59PM +0000, Andrew Ruder wrote:
On 10/20/2014 03:00 PM, Tom Rini wrote:
So, on new platforms we call dm9000_initialize with the right IO/DATA locations for the given device, yes? I think I'd rather update everyone else to call things the right and new way, rather than work-around supporting both.
The expectation is that new platforms would move over to dm9000_initialize_ex(), dm9000_initialize() just being a shim to use what used to be #define'd in the board config.h for backwards compatibility with older boards.
There's really 3 options that I fought with:
1.) Change dm9000_initialize() to dm9000_initialize(x, y, z). PATCH #3 then also includes changes to all of the various boards.
2.) Add dm9000_initialize_ex(x, y, z), make dm9000_initialize() call dm9000_initiailize_ex(x, y, z). No boards need to change. This is what I chose.
3.) Hybrid approach, do #2, make another patch (#4) that moves everything over to dm9000_initialize_ex(x, y, z) while renaming it to dm9000_initialize(x, y, z). Seems more round-about than #1 with the same end-result, but sometimes I feel like it is a little easier to review the meat of this change (#3) without it also dealing with tons of board churn.
Thoughts?
How about step 1 is checkpatch re-formatting only, step 2 is dm9000_initalize(x, y, z) but y/z aren't used, step 3 is passing around 'dev' and step 4 is the rest of the changes (so that y/z are used, if I follow all of the logic right). This splits the whitespace/etc churn out from everything else, then makes it easy enough to review that boards are converted right to the new logic.