https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123791
Bug ID: 123791
Summary: [16 Regression] Relocations against .rodata in
-fPIE/-fPIC code and other bugs on aarch64 since
r16-6776
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: jakub at gcc dot gnu.org
Target Milestone: ---
-O2 -fPIE or -O2 -fPIC -fvisibility=hidden
char *
foo ()
{
return (char *) (__UINTPTR_TYPE__) foo + 7483647;
}
regressed on aarch64-linux with
r16-6776-g1438f92c9e721133031b72a3884ed7bc17d0401c
We should never put constants into .rodata section for -fpic/-fpie if it
contains relocations.
default_elf_select_rtx_section does handle relocations right and also comdats,
even for those I think the r16-6776 change looks wrong, don't have a testcase
right now but I think putting rtxes into .rodata section (even with -fno-pic)
is undesirable because it needs to be removed by the linker if the particular
comdat group is not selected.
