I am putting the code into the SDRAM and let the processor execute the code from the SDRAM.
By default, the SDRAM controller (or Blackfin itself) seems to read a bunch of instructions (READ Burst) at a time and load them into the internal memory (L1) or cache so that the processor can run the code faster.
I want the processor directly execute the code from the SDRAM, meaning that the instructions are not loaded to cache/internal memory as a bundle (and how can I know how many instructions have been loaded? Since the READ Burst can be truncated anytime by other commands), but the processor reads one instruction, executes, and than reads another one, executes. How can I do this? Alternatively, if the SDRAM READ Burst feature can be disable, then I can know when the one certain instruction is being fetched and executed by watching the address bus and the CS, RAS, CAS and WE signals of the SDRAM.