In message OF59D7AF38.9E936776-ON07257066.0059D1CC-07257066.005B8D56@mck.us.ray.com you wrote:
As I understand it, these are the issues:
- Small stack
- BSS is uninitialized
- Global data is not writeable
Am I missing something from this list?
Many other parts are not yet initialized and/or working, and trying to change the init sequence to make it work will most probably result in a mess. You don't even have a serial console port yet for debugging. This also means that your have to remove each and every printf() or puts() from all code you need to call, and recursively.
That being said, please realize that some board architectures have *much more* resources available at boot than some of the powerQUICC systems that ppcboot/U-Boot grew up on.
I am definitely aware of this. However, the design decision to provide a serial console port very early also results in certain restrictions - like not being able to load the environment from arbitrary storage devices.
For example, the PPC405 systems that can be built using a VirtexII PRO FPGA have main memory (SDRAM, DDR SDRAM, etc) available *immediately* upon processor release. These FPGAs provide lots of flexible, internal static RAM (block RAM), which for a 2VP50 device is something like 400 kbytes. In fact, my current design has no FLASH at all - a large chunk of block RAM is initialized with the U-Boot image along with the rest of the FPGA logic. U-Boot is running from RAM from the start.
The U-Boot design is definitely *not* optimized for such a system.
There was a suggestion to use a dedicated partition on the CF card to hold the environment and manually parse the partition map. Is this an approach you could support?
All you get rid of is the file system problems.
You still have lots of other problems. I mentioned that you must "clean" all code you are going to call, and recursively all functions that are referenced by this code, from any printf() or puts(). Either you give up here, or you will have a mess of code. or you will try to delay access to the real environment and fake one, at which point the whole approach should be put into question.
Best regards,
Wolfgang Denk