Re: changing stacks

[Jeff Schwab]

Ganesh wrote:
> Hi,
>
> I need to switch the stack before a function call in C++. Can I just
> change the  stack pointer just before the function invocation. Is that
> all that is required to change stacks? Like,
>
> void func {
>   change_stack_pointer
>   foo (exp1, exp2...)
>   restore_stack_pointer
>
> }

Why are you doing that?  Forgive the question, but I've never seen 
anything like that in C++, and I'm curious.  Have you got one thread 
trying to call a function in the context of another?  If so, I have to 
wonder why you don't just have the correct thread call the function in 
the first place.

> The function call has expressions that access the local variable in
> the original stack.   I am concerned that if a compiler (say g++)
> emits a code that accesses the local variable of func using address
> relative to SP, then the above approach will fail (because the SP is
> changed).

That is a very reasonable concern.  Care was taken in the design of the 
C and C++ programming languages to make sure that you would never need 
to do anything like this, outside of very low-level scheduling code.

> How shall I gurantee in g++ (and if possible in other
> compilers) that access to local variables is not relative to SP? Or is
> it always the case that the access is not relative to SP?

It depends on the architecture, but local variable addresses on x86 are 
typically relative to the base pointer EBP, not the stack pointer ESP. 
The -fomit-frame-pointer flag allows GCC to use offsets directly from 
the stack pointer.

On x86, with C-style calling conventions, the register EBP is copied 
from ESP inside the function call.  This means that if you modify the 
stack-pointer before the function is called, EBP is copied to late to 
make a difference.  It will be copying a potentially invalid stack-pointer.

> I know that there are other ways to change stack (makecontext,
> sigaltstack etc). But, I want an extremely light weight mechanism.