Linux 2.4.20, x86 for testing, ARM target

What I want to do is this:

1. compile a .o file, save to disk (on command line).
This .o has a statically initialized function table pointing to the public functions in the .o file
e.g.

------- from file1.c -----------
typedef int MY_ADD(int, int);
int my_add(int left, int right) { return left + right; }

struct ftable { MY_ADD *ft_add; };

static struct ftable my_table = { my_add };
---------------------------------

2. The rest happens in a C program which is not linked to file1.o, but has declarations
for the needed types

3. open file1.o, copy to RAM

4. figure out where the data and object code resides in RAM

5. get a pointer to my_table

6. fix up the function pointer in my_table to (re)point to my_add

7. call my_add from my C program, e.g. my_table->ft_add(left, right);


Why not simply use a shared lib, kernel module, etc and let the kernel do the relocation? Because I want to make some code available to a booting kernel _before_ it has come up enough to load such things on its own.

I'm looking at the ELF format and can parse the headers and string tables. Before I delve into the symbol table, I'd like to see if there's a better/easier way.

Is there an objcopy switch(es) that would give me a raw binary image that would be better to parse than ELF?

Will I have problems from the memory manager since the code isn't necessarily loaded in an "ok to execute" area?

Thanks!

Welcome to LQ!
This post is more likely to be answered in the programming forum.