Bug 233233 - .gnu.hash support is incompatible/unsupported on mips
|
Bug#:
233233
|
Product: Gentoo Linux
|
Version: unspecified
|
Platform: MIPS
|
|
OS/Version: Linux
|
Status: RESOLVED
|
Severity: critical
|
Priority: P2
|
|
Resolution: FIXED
|
Assigned To: toolchain@gentoo.org
|
Reported By: kumba@gentoo.org
|
|
Component: Development
|
|
|
URL:
|
|
Summary: .gnu.hash support is incompatible/unsupported on mips
|
|
Keywords:
|
|
Status Whiteboard:
|
|
Opened: 2008-07-29 07:00 0000
|
With the addition of 77_all_generate-gnu-hash.patch, newer binutils will not
function properly on mips due to an explicit check for .gnu.hash attempts by
ld. If --hash-style=sysv is not explicitly set in make.conf via LDFLAGS or
CFLAGS, then the following error is encountered:
/usr/lib/gcc/mips-unknown-linux-gnu/4.3.1/../../../../mips-unknown-linux-gnu/bin/ld:
.gnu.hash is incompatible with the MIPS ABI
This is because .gnu.hash support is incompatible with the MIPS ABI, and very
likely, never will be compatible. The gory details are available in the
following message:
http://www.sourceware.org/ml/binutils/2006-07/msg00341.html
We need to either have mips force --hash-style=sysv via some means/mechanism,
or this patch needs to be skipped when binutils is compiled on a mips host or
being compiled for a mips host.
i've mentioned this to vapier several times but nothing has ever come of it.
forcing -Wl,--hash-style=sysv in the profile would work for the most part but
any package ignoring LDFLAGS will still break. i think the best way to do this
is to make 77_all_generate-gnu-hash.patch conditional on !mips.
Yeah, though considering all the patching code is in an eclass, I'm not sure
how a patch can be applied to all archs but one. I think the patch naming
scheme allows one to apply only to a specific arch, or all archs only.
hmm. one quick and dirty way i've done this before was to have something like
78_mips_read-my-mips_no-gnu-hashes.patch which is 77 reversed. there may be a
better way of doing it though.
what do you know, epatch takes arguments. :P
src_unpack(){
toolchain-binutils_src_unpack
use mips && EPATCH_OPTS="-R" epatch
"${WORKDIR}"/patch/77_all_generate-gnu-hash.patch
}
Yeah, but that means touching the ebuilds for binutils directly. Probably
easier to apply a reverse patch that fires for mips only, since the offending
patch is a one-liner.
*** Bug 232139 has been marked as a duplicate of this bug. ***
Doesn't look like the change to 77_all_generate-gnu-hash.patch works:
# mips64make -j3 vmlinux
CHK include/linux/version.h
CHK include/linux/utsrelease.h
Checking missing-syscalls for N32
CALL scripts/checksyscalls.sh
Checking missing-syscalls for O32
CALL scripts/checksyscalls.sh
CALL scripts/checksyscalls.sh
CHK include/linux/compile.h
LD usr/built-in.o
mips64-unknown-linux-gnu-ld: .gnu.hash is incompatible with the MIPS ABI
LD init/mounts.o
mips64-unknown-linux-gnu-ld: .gnu.hash is incompatible with the MIPS ABI
make[1]: *** [init/mounts.o] Error 1
make: *** [init] Error 2
make: *** Waiting for unfinished jobs....
make[1]: *** [usr/built-in.o] Error 1
make: *** [usr] Error 2
LD arch/mips/sgi-ip30/built-in.o
mips64-unknown-linux-gnu-ld: .gnu.hash is incompatible with the MIPS ABI
make[1]: *** [arch/mips/sgi-ip30/built-in.o] Error 1
make: *** [arch/mips/sgi-ip30] Error 2
The problem with that warning/error message out of the ld is the return code
triggers a failure to the make system, killing the compile. Removing the 77_
patch looks so far like the only thing that works for us.
i imagine it's only failing in the cross case. native prob works fine.
Native does work fine, but this essentially breaks crossdev for mips targets,
because the #ifdef __mips__ check passes on a non-mips system, and binutils
gets built with .gnu hash support. End result:
# mips64make -j3 vmlinux
CHK include/linux/version.h
CHK include/linux/utsrelease.h
Checking missing-syscalls for N32
CALL scripts/checksyscalls.sh
Checking missing-syscalls for O32
CALL scripts/checksyscalls.sh
CALL scripts/checksyscalls.sh
CHK include/linux/compile.h
GEN .version
CHK include/linux/compile.h
UPD include/linux/compile.h
CC init/version.o
LD init/built-in.o
mips64-unknown-linux-gnu-ld: .gnu.hash is incompatible with the MIPS ABI
make[1]: *** [init/built-in.o] Error 1
make: *** [.tmp_vmlinux1] Error 2
# uname -a
Linux khazad-dum 2.6.25.6 #4 SMP Sun Jun 22 14:21:56 EDT 2008 i686 AMD
Athlon(tm) MP 2000+ AuthenticAMD GNU/Linux
So I have to resort to tricks to get cross-binutils to build properly so I can
cross-compile mips kernels.
I just add this in toolchain-binutils.eclass for workaround in case of cross
compile
remove gnu-patch patch file for this target
--- /usr/portage/eclass/toolchain-binutils.eclass 2007-11-12
03:06:11.000000000 +0700
+++ /usr/local/portage/eclass/toolchain-binutils.eclass 2008-11-23
16:40:25.992977846 +0700
@@ -114,6 +114,10 @@
if ! use vanilla ; then
if [[ -n ${PATCHVER} ]] ; then
EPATCH_SOURCE=${WORKDIR}/patch
+ if [[ ${CTARGET} == mips* ]] ; then
+ # remove gnu-hash for mips (bug #233233)
+ rm -f ${EPATCH_SOURCE}/*gnu-hash*.patch
+ fi
[[ -n $(ls "${EPATCH_SOURCE}"/*.bz2 2>/dev/null) ]] \
&& EPATCH_SUFFIX="patch.bz2" \
|| EPATCH_SUFFIX="patch"
if we want to do it that way just set EPATCH_EXCLUDE.
This is related to #229173
No comments/complaints? I'll put this in in a few days unless otherwise. It's
holding us back on some things.
*** Bug 246573 has been marked as a duplicate of this bug. ***
*** Bug 229173 has been marked as a duplicate of this bug. ***
