PATCH: allow 2.6 to build on old old setups

PATCH: allow 2.6 to build on old old setups

[Alan Cox]

(Mikael Pettersson)
diff -u --new-file --recursive --exclude-from /usr/src/exclude linux-2.6.0-test2/scripts/modpost.c linux-2.6.0-test2-ac1/scripts/modpost.c
--- linux-2.6.0-test2/scripts/modpost.c	2003-07-10 21:05:39.000000000 +0100
+++ linux-2.6.0-test2-ac1/scripts/modpost.c	2003-07-23 15:37:13.000000000 +0100
@@ -296,12 +296,14 @@
 		/* ignore global offset table */
 		if (strcmp(symname, "_GLOBAL_OFFSET_TABLE_") == 0)
 			break;
+#ifdef STT_REGISTER
 		if (info->hdr->e_machine == EM_SPARC ||
 		    info->hdr->e_machine == EM_SPARCV9) {
 			/* Ignore register directives. */
 			if (ELF_ST_TYPE(sym->st_info) == STT_REGISTER)
 				break;
 		}
+#endif
 		
 		if (memcmp(symname, MODULE_SYMBOL_PREFIX,
 			   strlen(MODULE_SYMBOL_PREFIX)) == 0) {
\x05

Re: PATCH: allow 2.6 to build on old old setups

[David S. Miller]

On Sun, 27 Jul 2003 21:26:36 +0100
Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:

> +#ifdef STT_REGISTER
>  		if (info->hdr->e_machine == EM_SPARC ||
>  		    info->hdr->e_machine == EM_SPARCV9) {
>  			/* Ignore register directives. */
>  			if (ELF_ST_TYPE(sym->st_info) == STT_REGISTER)
>  				break;
>  		}
> +#endif

This change is wrong.

If you're going to do this, it's much better to define it to the
correct value in this case (which is decimal '13').

Re: PATCH: allow 2.6 to build on old old setups

[Alan Cox]

On Llu, 2003-07-28 at 02:52, David S. Miller wrote:
> >  		    info->hdr->e_machine == EM_SPARCV9) {
> >  			/* Ignore register directives. */
> >  			if (ELF_ST_TYPE(sym->st_info) == STT_REGISTER)
> >  				break;
> >  		}
> > +#endif
> 
> This change is wrong.
> 
> If you're going to do this, it's much better to define it to the
> correct value in this case (which is decimal '13').

Its sparc specific stuff so presumably all sparc stuff had the register
?. I can change and resubmit though - no problem

Re: PATCH: allow 2.6 to build on old old setups

[Mikael Pettersson]

Alan Cox writes:
 > On Llu, 2003-07-28 at 02:52, David S. Miller wrote:
 > > >  		    info->hdr->e_machine == EM_SPARCV9) {
 > > >  			/* Ignore register directives. */
 > > >  			if (ELF_ST_TYPE(sym->st_info) == STT_REGISTER)
 > > >  				break;
 > > >  		}
 > > > +#endif
 > > 
 > > This change is wrong.
 > > 
 > > If you're going to do this, it's much better to define it to the
 > > correct value in this case (which is decimal '13').
 > 
 > Its sparc specific stuff so presumably all sparc stuff had the register
 > ?. I can change and resubmit though - no problem

The error is that modpost.c is compiled against the C library elf headers
instead of the kernel's own elf headers. My #ifndef patch is just a workaround
for a missing -Iinclude and possibly -nostdinc when modpost.c is compiled.

David's response to that was that I should update my C library headers
instead. This is doable, but kind of defeats the purpose of having test
machines with old user-spaces around in the first place.

Linus once said that Linux did its own offsetof() just to avoid relying
on a fully working user-space...

Re: PATCH: allow 2.6 to build on old old setups

[David S. Miller]

On 28 Jul 2003 12:30:59 +0100
Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:

> On Llu, 2003-07-28 at 02:52, David S. Miller wrote:
> > >  		    info->hdr->e_machine == EM_SPARCV9) {
> > >  			/* Ignore register directives. */
> > >  			if (ELF_ST_TYPE(sym->st_info) == STT_REGISTER)
> > >  				break;
> > >  		}
> > > +#endif
> > 
> > This change is wrong.
> > 
> > If you're going to do this, it's much better to define it to the
> > correct value in this case (which is decimal '13').
> 
> Its sparc specific stuff so presumably all sparc stuff had the register?
> I can change and resubmit though - no problem

The situation is that, on every platform, one gets the define in
/usr/include/elf.h, it's just that older glibc's lack the define.