Haavard Skinnemoen wrote:
<snip>
Right...I'm beginning to doubt that anyone is familiar enough with the u-boot code, since everyone seems to have their own opinion about how things are supposed to work.
To summarize, here are the possible ways to fix the problem as I see it:
- Use virtual address in CONFIG_ENV_ADDR to conform with the way the CFI driver currently works. Rejected by Wolfgang because virtual addresses don't exist.
- Fix the API and user interface breakage by reverting commit 09ce9921. Rejected because virtual addresses are used everywhere.
- Fix it by using map_physmem() in a way that works for all architectures. Rejected because all other architectures than PPC are evil and need to be punished for doing things differently.
Your "triple revert" patch doesn't look overly complex, and seems to only add a few map_physmem() calls (I'm summarising *quite* a bit there !!).
Is there not some way using weak functions (or similar) to add some AVR32 specific workarounds.
Or ... there's *plenty* of arch specific #ifdefs in most of the rest of u-boot, so could we not just "#ifdef AVR32" the existing code for the time being until this sticking point gets unstuck ?
- Introduce a custom flash driver for ATNGW100. Rejected because stupid principles are more important than making things work.
So I don't really know where to proceed from here. I guess two additional options are forking the damn thing or creating a proprietary bootloader, but I don't really want to do either.
Me neither !!
Mark