_start call ABI

[chorman0773]

This does not need to be fully decided until we are at userspace, but it is probably a good idea to decide on this now.

Basically, how should we call the entry point of user applications, or of the interpreter when there is a PT_INTERP segment.

Obviously, we need to store the arguments on the stack before the function, and then an array of pointers to them. Should we then push a pointer to that array, or pass it in rdi/rsi (if so, do we pass argc first in rdi?). Should we pass environment variables in the same way, or provide a system call to access them.

I'd note that there really isn't an "incorrect" way of doing this, since we'd need assembly anyways to far jmp into user code with CPL=3.

[rdrpenguin04]

I'd prefer not passing environment variables, since that then requires macro hackery to get the environment variables cross-platform, besides the existing hackery required to use the variables.

[chorman0773]

Well, there needs to be a way to get environment variables for, at the very least, getenv.

[rdrpenguin04]

Yeah, I'm going to have to recommend a syscall. I definitely do not recommend passing environment variables to main. (or _start for that matter, since that itself encourages passing to main)

[chorman0773]

Any comment on the other two (argc and argv) being passed into start.

[rdrpenguin04]

I think that sounds good, as they're more likely than not going to be used.

[chorman0773]

Well, they need to be available to be passed into main. My question is if they should be in rdi and rsi (like Sys-V) or just store them on the stack like unix does and glibc's crt1.S handles.

[rdrpenguin04]

Probably put them on the stack so that initialization can do other things (which, it almost always will) without needing to specifically push rdi and rsi.

[chorman0773]

Apparently Sys-V specifies this as well, so we should probably conform with it.