Re: [PATCH v2 0/8] clean up vdso-layout.lds.S

[Petr Tesarik <ptesarik@suse.cz>]

H. Peter Anvin píše v Po 15. 06. 2009 v 11:33 -0700:
> Petr Tesarik wrote:
> >>>
> >> *(COMMON) is BSS, not data...
> > 
> > Very true, but for the vDSO we decided to put both writeable and
> > read-only data into one section (called .data for that matter), probably
> > to reduce the number of sections and hence also the size of the
> > resulting binary.
> > 
> 
> BSS is neither,

Right. My typing was once again faster than my thinking. I meant
initialized vs. uninitialized, of course.

>  but I guess for the vDSO there really isn't any such
> thing as uninitialized content.

Right, too. I can't even think of a valid use case, so .broken might be
the right place for both COMMON and .bss.

> I have to admit feeling funny about that, and I'm wondering if we
> shouldn't compile the vDSO with -fno-common.

Oops, sorry. It's already compiled with -fno-common, because it gets
inherited from the top-level Makefile's KBUILD_CFLAGS. Sorry for the
noise.

Anyway, my feeling is that the whole discussion is a bit academic. If I
want to be rigorous, I should take an opt-in approach, i.e. handle all
sections that work fine (e.g. also debugging sections) and then put all
else into .broken with something like:

.broken {
	/* All else is dubious. */
	*(*)
}

But this can become a maintenance PITA. Whenever GCC and/or binutils add
a new extension, the linker script would have to be adjusted
accordingly. Well, maybe that's even correct, because somebody at least
stops and thinks for a while about the implications of the new feature.
But I'm not really offering to become the new maintainer of this file.

Comments?

Petr Tesarik