RFC: Disallow protected data symbol with copy relocation?

[H. J. Lu]

On Sat, Jan 22, 2005 at 02:21:33PM +0100, Marc Espie wrote:
> In article <20050121064513.GA28057@lucon.org> you write:
> >I think it has something to do with the psABI and Linux runtime,
>                                                    ^^^^^^^^^^^^^
> >something like relocation is 32bit and main/DSO are several GB
> >apart. For x86_64, we can warn it when creating such a shared
> >library and refuse to link if it is referenced by main.
> 
> So, if I interpret this correctly, this doesn't work on Linux due to
> some specific aspects of the memory map and the way the dynamic linker
> works.
> 
> Do you have any idea how this translates to other OSes ?
> 

Linux uses more than 2GB address space on x86_64. I'd like to hear what
address arrange other x86_64 ELF OSes use. If they only use from 0
to 2GB for all applications, they don't have this issue.

> Am I misreading your proposal, or do I see blanket statements such as
> `ia32' or `x86_64', which do lead me to believe you want to force this
> for any ia32/x86_64 elf target, without any regard for the actual OS ?

I'd like to see what other OSes do. These are generic psABI problems.
I would love to hear solutions for them.


H.J.