Am 08.06.09 18:35 schrieb(en) Wolfram Sang:

quoted

Question: why is bank-width even relevant for a RAM device?

The underlying map_ram-driver uses it once while erasing. The  
question remains if this is really needed?

Am 06.06.09 18:16 schrieb(en) Albrecht Dreß:

At least if the RAM is attached to the 5200's Local Plus Bus in  
16-bit mode, no byte write accesses are allowed (byte /reads/ work,  
though).  I have a tweak (which I will post next week) to address  
this case, which depends upon this setting.

To put this clearer: on '5200 based systems, the driver (more specific:  
the function inline_map_copy_to()) *must* know whether the hardware is  
connected in 8-bit or 16-bit mode to the Local Plus Bus, as byte writes  
(issued by memcpy_toio()) will fail for the latter setup (probably the  
same applies for byte and word writes in 32-bit mode).

IMHO, this information should be passed using the device tree.  The  
"bank-width" seems to be an obvious choice for that.

Best, Albrecht.