It's done in linker script. I'm not sure how it works if you try to load the code directly to RAM, but it's usually done such that the code is loaded to flash and copied into RAM by the program. Then, in linker script, you need to use both load addresses and link addresses. Then the code is linked to be run from RAM, but loaded into flash.
Bit like this:
.text2 :
{
. = ALIGN(.,0x8);
*loader.o(.text)
*rpi2.o(.text)
*serial.o(.text)
*util.o(EXCLUDE_FILE(*instr_util.o).text)
*gdb.o(.text)
*(.text)
} >EXEC AT>LOAD
(From my rpi_stub)
Here the code is loaded to memory region LOAD, where most of the code is copied to memory region EXEC and run there. The copyer part of the program is not in this section.
IF you load directly on RAM, then you define a memory section in RAM and link the code there. Then you don't need the load address. It's just that restarting may not work. 🙂
