lib/csu/riscv: Remove separate assembly startup file

[arichardson]

Summary

There is no need to do the pointer arithmetic in assembly, when we
can do it in C instead. This removes the separate assembly file and
implements a tiny naked function that sets up gp and calls into C.

This ensures that the object file has all the required attribute set
without having to take care to include directives such as
.section .note.GNU-stack,"",%progbits, etc.
Computation of argv/envv is also understandable now without needing to know
RISC-V assembly code.

Test Plan

Still boots fine, godbolt confirms assembly matches

[jrtc27]

lib/csu/riscv/crt1_c.c

50–52

62

used isn't needed, attributes normally come after the return type

72

This constraint makes me nervous. GCC uses "l,S,U" for its call operand, whereas "i" is for immediates. I can't find documentation for what l and U are exactly, but S is "A constraint that matches an absolute symbolic address." Is there a reason not to just put the __start inline and mark it __used?

[arichardson]

lib/csu/riscv/crt1_c.c

62

I guess it is not needed since it is the entry point, just being paranoid since I've had a few cases lately where LTO+linker GC removed functions I assumed were used.

72

I recently came across the recommendation to use "i" in Arm documentation, I believe it was the following:

https://developer.arm.com/documentation/101754/0622/armclang-Reference/armclang-inline-assembler/Symbol-references-and-branches-into-and-out-of-inline-assembly?lang=en

This was rather helpful since I was using it to call a C++ function and it takes care of C++ name mangling. It works as expected on both Clang and GCC (even with old versions): https://godbolt.org/z/ejbjsT3Yv

[arichardson]

review comments

[arichardson]

lib/csu/riscv/crt1_c.c

50–52

I kept the parens here since they don't do any harm but can remove them if needed.