https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82501
--- Comment #25 from Andrey Drobyshev <a.drobyshev at samsung dot com> ---
(In reply to Martin Liška from comment #23)
> (In reply to Andrey Drobyshev from comment #22)
> > Created attachment 45851 [details]
> > Work-in-progress fix considering relocations
> >
> > I'm a bit stuck. I managed to precompute reloc value for the globals (the
> > code for detect_reloc_for_node() is stealed from get_variable_section ()). I
> > also succeded in putting dummy variable into .data.rel.ro by making a decl
> > of a pointer type.
>
> Nice progress!
>
> > But .data.rel is not working: checks from varasm.c:1173
> > and varasm.c:1197 are somewhat conflicting. Maybe you could find a way to
> > overcome this?
>
> Can you please paste content of the lines, the lines you mentioned point to
> a '}' for me.
Sorry for the delay, was out of the office for a couple of weeks...
What I actually meant was the following: setting DECL_INITIAL to
error_mark_node and making the decl of pointer type for computing the right
reloc here:
if (DECL_INITIAL (decl) == error_mark_node)
reloc = contains_pointers_p (TREE_TYPE (decl)) ? 3 : 0;
else if (DECL_INITIAL (decl))
reloc = compute_reloc_for_constant (DECL_INITIAL (decl));
else
reloc = 0;
along with making the decl non read only (as we want to put it in the data.rel)
causes it to be put in .bss here (in particular, because bss_initializer_p
(decl) becomes true):
if (ADDR_SPACE_GENERIC_P (as)
&& !DECL_THREAD_LOCAL_P (decl)
&& !(prefer_noswitch_p && targetm.have_switchable_bss_sections)
&& bss_initializer_p (decl))
{
if (!TREE_PUBLIC (decl)
&& !((flag_sanitize & SANITIZE_ADDRESS)
&& asan_protect_global (decl)))
return lcomm_section;
if (bss_noswitch_section)
return bss_noswitch_section;
}
