Right, here's a working solution for putting ARM code in WRAM:

Use a simple linker script which looks something like this:

OUTPUT_FORMAT("elf32-littlearm", "elf32-bigarm", "elf32-littlearm")
OUTPUT_ARCH(arm)
ENTRY(_start)

MEMORY
{
iram (wx) : ORIGIN = 0x3000000, LENGTH = 32K
eram (wx) : ORIGIN = 0x2000000, LENGTH = 256K
rom (rx) : ORIGIN = 0x8000000, LENGTH = 8M
}

SECTIONS
{
.bss :
{
*(.bss)
}
> iram

.text : AT(0x8000000)
{
*.o(.text)
*(.data)
*(.rodata)
*(.glue_7t)
*(.glue_7)
}
> rom

.armcode : AT(ADDR(.text) + SIZEOF(.text))
{
_sdata = .;
*.ao(.text);
_edata = .;
}
> iram
_datalma = LOADADDR(.armcode);

PROVIDE (end = .);

.stab . (NOLOAD) :
{
[ .stab ]
}
.stabstr . (NOLOAD) :
{
[ .stabstr ]
}
}

This puts all .o files in the ROM, including all code and data (Sorry, no
initialised global support. You have to do that yourself...).

It also puts all .ao files in work RAM after the global variables (BSS).
These are you ARM files! Make sure you compile all ARM code to .ao object
files for this to work...

All you need to do now is copy the code to work RAM using something like
this in your crt0.s (nicked from Jeff's code):

@-- COPY INITIALISED DATA / CODE TO RAM

ldr r1,=_edata
ldr r2,=_sdata
subs r3,r1,r2 @ Is there any data to copy?
beq CIDSkip @ no

ldr r1,=_datalma
CIDLoop:
ldmia r1!,{r0}
stmia r2!,{r0}
subs r3,r3,#4
bne CIDLoop

CIDSkip:

Then jump to your main here...

That's it!

Oh, except GCC is crap at working out how to call stuff, so you have to call
ALL your ARM functions via function pointers. Likewise if you want to call
THUMB function from the ARM code you have to use a function pointer.

Not sure about the ARM compiler. Will look into it later... :)

Cheers,

--Jim