Wolfgang Denk a écrit :
Dear Albert ARIBAUD,
In message 4B1A05A9.4040104@free.fr you wrote:
No, this is not correct. get_ram_size() is used always on a single bank of memory only.
Do you mean calling get_ram_size() four up to times based on the configured number of banks and configured sizes? I then fail to see the
Yes, exactly.
added value of get_ram_size() wrt using the configured sizes directly.
The added value is that get_ram_size() will detect (1) a lot of common error situations and (2) will detect the actual size of the respective memory banks.
Assume you have a system where differentt types of memory chips can be fit, or where you can insert memory modules. Then you configure for the largest possible type, and get_ram_size() will detect what's really present, so you can adjust the configuration. See the README for details.
However I realize that this code is actually SoC-specific, not board-specific. It could be moved in cpu/arm926ejs/orion5x/dram.c, and then orion5x_sdram_{bar,bs} could be made static (or inlined).
What do you think?
Seems to make sense.
All right. Does doing that lift the requirement to use get_ram_size()?
I see no reason yet why you would not want to use get_ram_size() - the memory test feature alone is useful enough, even if you never intend to use different RAM sizes.
Ok. Considering the machine would have had fixed sized banks but the SoC would not, get_ram_size() indeed makes sense especially if I move the code to cpu/arm926ejs/orion5x/dram.c. I'll do that in V4 of the patch, once I get complete feedback for V3.
Amicalement,