In message 637EA992-FC18-11D8-9008-000393C7979E@student.unsw.edu.au you wrote:
I've had two problems. The first has been that that the code uses the get_timer(x) function, which is not implemented properly on ARM. (It
Indeed, it seems to be broken.
just returns the value of the OS counter, ignoring the value passed to it. I had to replace the get_timer stuff with get_timer_masked and reset_timer_masked. I could try fixing the arm implementation of get_timer, but figured you guys wouldn't want me touching that bit of code since other bits and pieces now depend on it. Is there some way of doing this nicely?
There should be no software depending on the broken behaviour. If there was any, it must be fixed, too.
The second was that one part of the code munges some bytes into a word to get them ready for writing. This is broken with respect to endian-ness, and breaks on my XScale running in little-endian mode. Is it possible (or appropriate?) to add a CFG_LITTLE_ENDIAN (or something similar)? Is there a better way of switching between the two? It would be nice if this code could be re-used without these sorts of problems.
If you use code from a board directory, no such endianess support is needed, as the board will normally not change endianness (at least I don't know of any of the supported boards which can be configured both for BE or LE).
Endianess is considered an issue only in case you're using a generic driver (like drivers/cfi_flash.*); but AFAIK the cfi_flash code already deals properly with the target byteorder.
Best regards,
Wolfgang Denk