On 09/26/2010 09:57 AM, Steve Sakoman wrote:
On Sun, 2010-09-26 at 16:37 +0200, Wolfgang Denk wrote:
Dear Nishanth Menon,
In message4C9F3E4F.3040403@gmail.com you wrote:
I think you are referring to SDRAM initialization. Yes, it is assumed that SDRAM initialization(at least the DMM part of it) is done before the call to sdram_size(). The right location for this seems to be the initial part of sdram_init().
thanks for the clarification.. NOR boot is usually going to have initial part in XIP -> we'd traditionally relocate to SDRAM before the call to sdram_size() happens.
Yes, this was the way how many ARM boards id, and this is what I'm trying to explain: this is WRONG.
SDRAM configuration must be done while still running from NOR, i. e. before relocation. Only then free reconfiguration, auto-sizing etc. is possible.
I'll make sure that this is done properly when(if?) a patch series for an OMAP4 board with NOR boot is prepared.
Getting back to the subject patch, though :-)
This patch fixes a bug in the current Panda board support in rc2. Without this patch U-boot will return an erroneous size for available RAM. We should consider it for inclusion as soon as practical.
I tend to agree, I might rather have my panda functioning properly and improve the location of sdram_size read at a later point of time -> for cleanups, we should probably start with the mux stuff..
That said, can we compromise by adding a FIXME: comment so that we dont really forget to fix it later on?
Regards, Nishanth Menon