I was wondering if someone with experience with compilers could tell me if my idea is workable or not.
Basically, I am going to propose a project idea to one of my professors and take a "project course" – I won't detail the process here much, just saying that its an undergrad project course where I get to propose the idea.
I am going to make a new RISC ISA (similar to MIPS but without delay slots, no floating point ...) and write a software emulator for it, as this is to be done entirely in software.
But I am going to make it a Harvard architecture CPU – meaning, data can not be executed; code and data are kept separate.
Making the ISA and an emulator for it is relatively easy, I don't expect to run into anything that I am not already familiar with.
Then I want to make a gcc or llvm backend for my new ISA so that C programs can be compiled for my new ISA.
Now I have never written a compiler. Given that my ISA is mostly modelled after MIPS, I can just model the backend after the MIPS (or some other RISC) backend.
The question I have is about the Harvard architecture part. Should I expect to run into major problems here? How would this complicate the code generation part?
In the end, I will write a report on my emulated Harvard CPU and analyse some security aspects of it (i.e. not allowing data execution could prevent buffer overflow attacks etc., etc. ...).
In case I abused the term "Harvard architecture" here, let me clarify what I mean. I am not talking about internal caches or anything like that. I am just referring to having the CPU keep code and data separate so that data can not be executed.