[U-Boot] [PATCH] [RFC] memsize.c: adapt get_ram_size() for address spaces >32 bit

[Wolfgang Denk]

Dear Wolfgang Wegner,

In message <20100527185909.GG31271 at leila.ping.de> you wrote:
> 
> as the systems I was working on always had far less memory, I can
> not comment much on the extension you propose here, but as Timur
> Tabi's comments seem to go in a direction which could lead to a
> bigger overhaul/discussion, I would like to add 2C from my point
> of view on coldfire...
> 
> - MCF53xx/MCF5445x both simply lock up if non-existent memory is
>   accessed. So if the SDRAM controller is set up for a too big size
>   of memory, get_ram_size() will fail. I assume this applies to
>   most coldfire devices.

The design of get_ram_size() is based on the assumption that you start
with a "big enough" mapping. Usually we map twice the maximum possible
size that actually can be fit.

> - How about systems/configurations using CONFIG_MONITOR_IS_IN_RAM?

Here you obviously cannot use this. What some ARM systems are doing
right now is plain wrong. get_ram_size() must always and only be
called _before_ the code is running from RAM.

> - at least in the coldfire world, CONFIG_SYS_SDRAM_SIZE is quite
>   often used for cache setup in the assembly code. This contradicts
>   changing/setting the SDRAM size at runtime...

You may want to change this :-)


Note that we have plans underway to rework the ARM architecture to do
proper relocation including auto-sizing of the RAM.  Other
architectures are invited to follow.

Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Never underestimate the bandwidth of a station wagon full of tapes.
                                -- Dr. Warren Jackson, Director, UTCS