Relocatable executables (i.e. TSRs)

[thejpster]

The -C relocation-model=pie option for rustc will produce an executable where code and data are designed to be in the same region (i.e. RAM) and can be located anywhere provided their relative offsets remain the same.

This means we should be able to load a TSR if it has been compiled as pie. It could occupy the top part of RAM and hook itself into the OS, perhap by replacing one of the OS API functions with itself. This would require the OS API table to be in RAM, but that's plausible - it's not too large and we do have 4K of OS reserved RAM. It would also allow the loading of multiple programs simultaneously (e.g. running a linker during a compilation) without moving the first program out of the way so the second can have access to the usual start address (0x2000_1000).

[thejpster]

This is distinct from the "ropi-rwpi" model, where the RW data can be relocated separately to the RO code/data, and register R9 is used to tell the RO code where the RW data has been located. That doesn't apply here because we have the usual (for a microcontroller) case of code + data being together in RAM.

[thejpster]

NuttX has a relocating loader we can look at.

[thejpster]

Note, the first comment is wrong. PIC and PIE produce ELF files containing relocations - code that must be adjusted for its run address at load time. We'd need to change the neotron-loader to perform these relocations.