I know that in x86_64, there exists a 128 byte red zone above (or below, address-wise) the stack pointer that functions can use without subtracting from rsp.
This sounds to me like the only things that need to be aware of the red-zone are situations in which code will get interrupted by other code running on the same stack.
The only cases where this seems relevant to me are:
- OS-level interrupt handlers (either switch stacks without touching the red zone, or subtract red zone size before using the stack)
- user-level signal handlers (since AFAIK POSIX signals deliver on the same stack if sigaltstack() wasn't used)
In those cases, when that code handles the existence of a red zone, linking together code compiled with -mno-red-zone and without it should not be a problem, right?
It'd only be a problem if the code assumes -mno-red-zone and some code uses it anyway.
Is my reasoning about this correct?
If this is correct, why do so many OS-dev resources tell you to enable -mno-red-zone? is it just to reduce complexity?
