I am using the Nios Gnupro tools on an embedded design. Using some default build script, the code is compiled & linked to execute SRAM but is stored in Flash. The Flash image also has a small copy routine that copies everything to SRAM on boot and hops over.
So staring with the same srec file, I can either use GDB to download direct to SRAM and step thru the execution, or use a utility to burn the image (plus copy loop) to Flash. Standard embedded stuff.
Either way, what is strange is that while running, the mainline code is clearly being executed from SRAM, but every function call (or ISR) jumps from an SRAM address to a Flash address instead of remaining in SRAM.
The objdump output and nm output show only SRAM addresses for all functions. The code was linked with the -Ttext flag with the SRAM base. Aside from the short Flash copy routine (that is only part of the Flash image), the SRAM image should not even know about Flash, so how the heck is it getting there?
There has got to be something in the build script that is linking incorrectly or using the wrong libraries. Other ideas?