Hi,
Wolfgang Denk wrote:
Dear Scott Wood,
In message 4A034B09.7030105@freescale.com you wrote:
Or what replaced the "immr" structs?
The device tree, mainly...
Right, of course.
... But #defines can work for u-boot.Of course they _can_ work. But they can easily fail (as we just see in this patch), and we don't have typechecking. So until DT's are omnipresent, let's use structs in U-Boot, please.
You *do* have typechecking as long as the individual blocks are described with structs.
We could take immap to extremes by defining one big 4GiB struct that shows where memory, immr, flash, desired PCI bars, FPGAs, etc. are -- but that would be silly. IMHO, so is doing it at the immr level. :-)
How would you deal with blocks being at different locations in different chips? It's a lot easier to ifdef (or have the config file specify) a couple addresses than to ifdef the locations of fields in a struct, especially when you have more than a couple variations.
For what its worth, I'm with Scott here. Structures for register blocks is very nice and should be mandated and it seems they are maintainable. Locations of individual blocks (or number of incarnations thereof) seem to change frequently and thus tend to be less friendly to "whole internal address space" structures. So the latter may better be mapped by single defines. The correctness of them is easily validated and an incorrect value will immediatley be discovered.
Cheers Detlev