On Tuesday 29 November 2011 02:00:38 Wolfgang Denk wrote:
Sridhar Addagada you wrote:
Here is a scenario, i have a working version of the u-boot on my board, and am trying my changes. Am not sure it is going to freeze the board, so would like the present version to load the new version for testing before it can be burned to flash. Is it possible to do this or is it already done?
In the strict sense, this usually cannot be done - one of the most critical parts (that most oftenly is incorrect and thus bricks your system) is the low-level initialization, like initializing the memory controller. And you usually cannot re-do this while already running from RAM.
Note that this is covered in the FAQ: http://www.denx.de/wiki/view/DULG/CanUBootBeConfiguredSuchThatItCanBeStarte dInRAM
i like this one: Answer: Well, yes. Of course this is possible. This is software, so *everything* is possible.
not to keep harping, but the Blackfin arch does support this. and does so stable-ly/cleanly imo. the cpu offers a few advantages though that most cores don't (afaik) which makes it possible: dedicated on-chip L1 SRAM that can be directly accessed (which is not cache). and since the majority of the code we have to have anyways for normal booting, the additional "i'm being booted from external RAM" logic adds like ~6 insns of overhead. so when we detect we're in RAM, we copy the initcode to on-chip memory, put external memory into self refresh, reprogram everything, take external memory out of self refresh, and then resume execution from RAM.
only caveat i know of is that you can't change some addressing modes (like # of cols used to program the SDRAM chips). otherwise, this changes how bytes are laid out in RAM, so you end up having stride-ed data. what was: 0xab 0xcd 0xef 0x00 ... is now something like: 0xab <garbage> 0xcd <garbage> 0xef <garbage> 0x00 <garbage> (the specifics here are probably wrong, but you get the general idea)
in this case, we aren't about to kick off a DMA to re-pack the data, so that is unsupported. but we find that's significantly less of an issue compared to what people actual want: to play with memory/clock timings.
</braindump> -mike