https://sourceware.org/bugzilla/show_bug.cgi?id=33552
Bug ID: 33552
Summary: gas --32 -mrelax-relocations=no sometimes generates
R_386_GOT32X
Product: binutils
Version: 2.46 (HEAD)
Status: NEW
Severity: normal
Priority: P2
Component: gas
Assignee: unassigned at sourceware dot org
Reporter: ro at gcc dot gnu.org
CC: hjl.tools at gmail dot com
Target Milestone: ---
Target: i?86-*-*
I've recently worked on a patch to always default to -mrelax-relocations=no
on Solaris/x86. For that, I meant to create a testcase showing that gas
doesn't generate relax relocations even without -mrelax-relocations=no.
However, I found that this options doesn't live up to it claims: starting from
testcase in gcc/configure.ac (HAVE_AS_IX86_GOT32X)
.data
bar:
.byte 1
.text
.global _start
_start:
cmpl $0, bar@GOT
jmp *_start@GOT
I found that the test sets HAVE_AS_IX86_GOT32X to 1 even though gas effectively
defaults to -mrelax-relocations=no on Solaris/x86.
$ gas --32 -mrelax-relocations=no got32x.s -o got32x.o
$ readelf -r got32x.o
Relocation section '.rel.text' at offset 0xa8 contains 2 entries:
Offset Info Type Sym.Value Sym. Name
00000002 00000103 R_386_GOT32 00000000 bar
00000009 0000022b R_386_GOT32X 00000000 _start
which was totally unexpected and contrarary to what the option is documented to
do. Exactly the same happens with trunk gas on Linux/i686 and has been this
way since binutils 2.26.1 when the option was introduced.
Looking closer, the option reliably inhibits the generation of
R_AMD64_GOTPCRELX
and R_AMD64_REX_GOTPCRELX for 64-bit, while for 32-bit it's only partially
effective. Running the gas --32 -mrelax-relocations tests with explicit
-mrelax-relocations=no shows just like the example above that out of those
tests
the gas/i386/got.s one shows the same bug.
--
You are receiving this mail because:
You are on the CC list for the bug.
