In article <yf9d6gzhdnv.fsf@mc.pp.se>, Marcus Comstedt <marcus@mc.pp.se> wrote:
>
>Hello.
>
>I'm trying to fix ld.elf_so for the sh3 port, but I have become rather
>puzzled by a discovery in the architecture independent rtld.c.
>
>At the beginning of the _rtld function is a comment which says
>
>     /*
>      * On entry, the dynamic linker itself has not been relocated yet.
>      * Be very careful not to reference any global data until after
>      * _rtld_init has returned.  It is OK to reference file-scope statics
>      * and string constants, and to call static and global functions.
>      */
>
>This is correct.  Referencing a global variable before _rtld_init
>gives a segfault.  However, what do we find just before the call to
>_rtld_init?
>
>        _rtld_pagesz = (int)pAUX_pagesz->a_v;
>        _rtld_init((caddr_t)pAUX_base->a_v, (caddr_t)relocbase);
>
>_rtld_pagesz is what?  A global data.  So it can't be referenced
>here.  And sure enough the linker crashes accordingly.  What gives?
>With bugs like this in the architecture independent code, does dynamic
>linking work on _any_ architecture?

This is definitely a bug...  I don't even think that _rtld_pagesz
is needed at that point, so fixing the bug is just swapping the
order of the two lines.

christos