[PATCH 2/2] arm/cpu/start.c: Avoid copying device-tree when possible

[Andrey Smirnov]

>> Oh, I don't think I mentioned in my previous e-mail, but I do have a
>> working Barebox image for that case. The way I have it implemented
>> right now is a vanilla, single board, no-PBL, no-relocation, i.MX6Q
>> SabreSD Barebox image with a minimal configuration. The only things I
>> had to change was device tree file -- default required to much RAM to
>> instantiate, so I had to trim it down -- and this patch to avoid
>> copying FTD that is already built-in. Oh, and I also had to disable
>> MMU, because page table takes about 1MB or RAM(I haven't had a chance
>> to spend any time trying to modify MMU code to support coarser
>> 1MB-page page table).
>>
>> The image is intended to be used by EEs to do DRAM related
>> experiments, so I do need a shell and that was the reason I went with
>> full Barebox instead of trying to cram it in PBL.
>
> Ok, I see. I think your original patch is almost fine, only the test if
> you need to copy or not needs adjustment. You have to test if the fdt is
> membase < dtb < membase + memsize. If it is you have to copy it,
> otherwise it should be fine to use it in place. Would that work?

Sorry for dropping this conversation. Unfortunately no, I don't think
this would work since in my case 'membase' is set to start of IRAM and
'memsize' is IRAM's size and since I am placing the image in IRAM that
test would come positive and the code would try to copy. What if the
condition was set to !RELOCATABLE && !PBL_IMAGE, so that if we know if
there's no PBL and no relocation we do not copy the data? Would that
be acceptable?