When building with BR2_REPRODUCIBLE the toolchain wrapper passes
-fdebug-prefix-map for all packages that are built. But this doesn't
affect the target libraries (like libgcc) built by GCC's build system.
GCC 4.3 added a configure option to set the debug prefix map for these
libraries, which is used here to avoid encoding potentially
non-reproducible build paths into the debug data.
Signed-off-by: John Keeping <john@metanate.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Commit "50332a530b gcc: rename option for ARC gcc" tried to add legacy
handling but the new symbol is part of a choice, and Kconfig does not
enforce the select of a option from a choice.
Update the legacy entry for 2016.11, following the example described in
the beginning of the file.
Cc: Giulio Benetti <giulio.benetti@benettiengineering.com>
Cc: Romain Naour <romain.naour@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Ricardo Martincoski <ricardo.martincoski@gmail.com>
Signed-off-by: Peter Korsgaard <peter@korsgaard.com>
This commit backports the patch "fixinc: don't "fix" machine names in
__has_include(...)" from upstream GCC, which is needed to resolve a
header conflict between glibc headers and kernel headers, which has
appeared since we bumped glibc to version 2.36 in commit
80c8c15c85.
The problem comes from the "fixinc" logic used by gcc to fixup some
headers files, generated inside an include-fixed/ folder. This logic
ended up replacing "linux/mount.h" by "__linux__/mount.h" in
__has_include() invocation, like this:
#ifdef __has_include
# if __has_include ("__linux__/mount.h")
# include "linux/mount.h"
# endif
#endif
in
build/host-gcc-final-11.3.0/build/gcc/include-fixed/sys/mount.h. With
this fix in place, this "include-fixed" header is no longer generated,
avoiding the problem.
This issue was visible in two different ways in glibc configurations:
- As a build failure during the gcc build itself, for architectures
that support libsanitizer, as libsanitizer includes mount.h, and
would therefore encounter the header conflict.
- As a build failure during another user-space package (such as
sysvinit for example), on architectures when libsanitizer isn't
used, and therefore for which the gcc build was successful, but the
header conflict shows up when building some "random" user-space
package.
The problem is already fixed in GCC 12.2.0, so no patch is
required. The problem did not exist back in GCC 8.4.0, so this version
does not need patching. Consequently, the patch is only needed for GCC
10.4.0, GCC 11.3.0 and the special ARC 2020.09-release version.
Fixes:
(gcc build issue, on architecture that supports libsanitizer)
http://autobuild.buildroot.net/results/90fe4c3b8b72a2c28555674383de9bbd9e8ae09a/
(sysvinit build issue, on architecture that does not support libsanitizer)
http://autobuild.buildroot.net/results/d7bf5795b7621a92be32f18794e3e67944fb96db/
(crun)
http://autobuild.buildroot.net/results/e3e8da4f797dced48aedf8c636db983d36849850/
(libarchive)
http://autobuild.buildroot.net/results/9fcbf0c036a97b2e9a4fcc6e173bcfa09e1b3dac/
Thanks a lot to Peter Seiderer for pointing the relevant GCC commit.
Fixes:
https://bugs.busybox.net/show_bug.cgi?id=15021
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Reviewed-by: Romain Naour <romain.naour@smile.fr>
Acked-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
package/gcc/11.3.0/0005-rs6000-Improve-.machine.patch:4: generate your patches with 'git format-patch -N'
package/gcc/11.3.0/0006-rs6000-Do-not-use-rs6000_cpu-for-.machine-ppc-and-pp.patch:4: generate your patches with 'git format-patch -N'
Signed-off-by: Arnout Vandecappelle <arnout@mind.be>
gcc 11.3.0 contains a backported patch [1] that introduce
a regression for old powerpc cpus like the powerpc 7400 (G4).
The glibc crash the init process due to a wrong asm machine
directive (.machine).
Run /sbin/init as init process
init[1]: segfault (11) at 7369693e nip 6f6e08 lr 6f6a68 code 1 in libc.so.6[690000+18f000]
init[1]: code: 280a000c 41c1ffe0 811edb80 554a103a 7d48502e 7d4a4214 7d4903a6 4e800420
init[1]: code: 2c08007a 4bffffbc 89290000 5529103a <7d2a482e> 2c090000 41c2ff78 7fe4fb78
Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
Backport two patches from the gcc-11 stable branch (the upcoming gcc
11.4.0).
[1] https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;h=3cb53c10831be59d967d9dce8e7980fee4703500
Fixes:
https://gitlab.com/kubu93/buildroot/-/jobs/2976071284
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Cc: Joel Stanley <joel@jms.id.au>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Currently, this option doesn't do anything. It only adds
--enable-plugins --enable-lto to the configure flags, but doesn't
disable them if it is not set. Since both of these default to enabled,
plugins and lto are effectively always enabled.
There really is no need to make this configurable: it adds a bit of size
and build time to host-gcc, but we don't care about that for host tools.
It's still up to individual builds to enable the LTO options.
Therefore, remove the option entirely. For clarity, explicitly pass
--enable-plugins --enable-lto to configure.
No legacy handling is added for the removed option. Since the behaviour
hasn't actually changed (independently of whether the option was enabled
or not), there's no point bothering the user with a legacy option.
elf2flt was linking with libdl depending on this option. Since the
option doesn't do anything, this is probably not needed. Still, to avoid
breaking things, and because linking with libdl doesn't cost us anything
anyway, always link with libdl.
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Currently, this option doesn't do anything. It only adds
--enable-plugins --enable-lto to the configure flags, but doesn't
disable them if it is not set. Since both of these default to enabled,
plugins and lto are effectively always enabled.
There really is no need to make this configurable: it adds a bit of size
and build time to host-binutils, but we don't care about that for host
tools. It's still up to individual builds to enable the LTO options.
Therefore, remove the option entirely. For clarity, explicitly pass
--enable-plugins --enable-lto to configure.
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
The toolchain for powerpc spe can use uClibc-ng without thread support.
So we need the same fix as commit [1].
[1] fff68f75b3
Signed-off-by: Romain Naour <romain.naour@smile.fr>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
gcc 12.1 is around, gcc 11.3 is the default version, so drop
9.5 in order to reduce the gcc choice.
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Even if gcc 10.x is still maintained for some time, switch to gcc 11.x
since it has been released since 2021-04-27 and gcc 12.x is available
since "2022-05-10".
We have been having toolchains in the autobuilders with gcc 11.x since
mid-June 2021, so the vast majority of the problems should have
already been solved.
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
libsanitizer has been enabled for mips64{el} in gcc 12 [1] but it
fail to build when n32 ABI is used:
In file included from output/mips64el-buildroot-linux-gnu/sysroot/usr/include/bits/stat.h:25,
from output/mips64el-buildroot-linux-gnu/sysroot/usr/include/fcntl.h:78,
from ../../../../libsanitizer/sanitizer_common/sanitizer_linux.cpp:55:
output/mips64el-buildroot-linux-gnu/sysroot/usr/include/bits/struct_stat.h:190:8: error: redefinition of ‘struct stat64’
190 | struct stat64
| ^~~~~~
In file included from ../../../../libsanitizer/sanitizer_common/sanitizer_linux.cpp:49:
output/mips64el-buildroot-linux-gnu/sysroot/usr/include/asm/stat.h:52:8: note: previous definition of ‘struct stat64’
52 | struct stat64 {
| ^~~~~~
Disable libsanitizer for mips64 with n32 ABI.
Note: Only glibc toolchains are affected since libsanitizer is
disabled for musl and uClibc-ng toolchains [2].
Fixes:
https://gitlab.com/kubu93/toolchains-builder/-/jobs/2510178651
[1] https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;h=344e6f9f2abcff9b2bb4b26b693be4a599272f43
[2] https://git.buildroot.net/buildroot/commit/?id=5f4d658d888b539de9a6247ae5b1a0999de5d4ec
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
When BR2_TOOLCHAIN_HAS_LIBQUADMATH is set, --enable-libquadmath-support
option is missing. So the float128 support is not fully enabled in gcc.
This lead to a build issue with gcc 12 on PowerPC power8 due to missing
M_2_SQRTPIq definition (provided by libquadmath.h).
../../../libgfortran/intrinsics/erfc_scaled.c: In function ‘erfc_scaled_r17’:
../../../libgfortran/intrinsics/erfc_scaled.c:143:22: error: ‘M_2_SQRTPIq’ undeclared (first use in this function); did you mean ‘M_2_SQRTPIf’?
143 | # define _M_2_SQRTPI M_2_SQRTPIq
| ^~~~~~~~~~~
This is fixed by adding --enable-libquadmath-support (like crosstool-ng
handling [1]).
Fixes:
https://gitlab.com/kubu93/toolchains-builder/-/jobs/2510178766
[1] https://github.com/crosstool-ng/crosstool-ng/blob/crosstool-ng-1.25.0/scripts/build/cc/gcc.sh#L370
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
This will use gcc-ar, gcc-nm and gcc-ranlib instead of the
normal binutils tools. The difference is that with the
wrappers, gcc plugins will be automatically picked up,
which is necessary to build with LTO.
With this enabled, it is possible to build everything (including libgcc
and libstdc++) with LTO by setting BR2_TARGET_OPTIMIZATION="-flto".
Note that you'd expect that the GCC build system would automatically do
this when --enable-lto is set, but this is not the case. There are some
open bugs [1][2] to allow building libgcc and libstdc++ with LTO support
but it's apparently not done yet.
Note that there are also reports of problems building libstdc++ with LTO
[3], but it seems that's no longer a problem (and the bug didn't get
updated).
Signed-off-by: Norbert Lange <nolange79@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=59893
[2] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77278
[3] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=60160
When the gcc arc version was bumped to a version using gcc
10.x (arc-2020.09-release) in commit 0791abfba0 (toolchain: update ARC
tools to arc-2020.09-release), the select of BR2_GCC_VERSION_ARC on the
appropriate BR2_TOOLCHAIN_GCC_AT_LEAST_xyz was not updated.
Commit 0b4c7ba01c (toolchain: update option descriptions for ARC tools
arc-2020.09-release) fixed the prompt, but still forgot to update the
appropriate BR2_TOOLCHAIN_GCC_AT_LEAST_xyz.
This commit eventually fixes this issue.
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
At the moment of gcc 11.1.0 release the OpenRisc patches for -mcmodel=large
were still pending. They have been upstreamed yesterday as pointed in gcc
bugzilla[1]. So they will be part of gcc 11.3.0 or maybe before on 11.2.
2. Anyway at the moment if we try to build packages libgeos and protobuf
with OpenRisc gcc 11.1.0 it fails due to missing -mcmodel=large. So let's
add OpenRisc patches for it as done for all the previous versions.
Fixes:
still not appeared on autobuilers
[1]: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99783
Signed-off-by: Giulio Benetti <giulio.benetti@benettiengineering.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
gcc 11.1 is around, gcc 10.2 is the default version, so drop
8.4 in order to reduce the gcc choice.
For PowerPC SPE, introduce BR2_GCC_VERSION_POWERPC_SPE
to avoid removing defconfigs arcturus_ucp1020_defconfig,
freescale_p1025twr_defconfig and qemu_ppc_mpc8544ds_defconfig
that are still used by some users [1].
BR2_GCC_VERSION_POWERPC_SPE use the same gcc version (8.4), no
change expected.
[1] 96e80ad214
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Cc: Oleksandr Zhadan <oleks@arcturusnetworks.com>
Cc: Michael Durrant <mdurrant@ArcturusNetworks.com>
Cc: Matthew Weber <matthew.weber@collins.com>
Cc: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Even if gcc 9.x is still maintained for some time (gcc 9.5 will be the
last), switch to gcc 10.x since it has been released since 2020-05-07
and gcc 11.x is available since 2021-04-27.
We have been having toolchains in the autobuilders with gcc 10.x since
mid-January 2021, so the vast majority of the problems should have
already been solved.
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
Let's add upstream patches introducing -mcmodel=large or1k gcc option that
works in conjunction with previous binutils patch. That option fix binutils
bug 21464[1] allowing to build libgeos with no problem. This way we can
consider buildroot toolchain binutils bug 21464 free.
[1]: https://sourceware.org/bugzilla/show_bug.cgi?id=21464
Signed-off-by: Giulio Benetti <giulio.benetti@benettiengineering.com>
Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be>
[Arnout: remove the PATCH M/N parts - cfr. check-package]
As reported on IRC by sephthir, the qemu_sparc_ss10_defconfig doesn't
work as expected: the system generated when booted under Qemu produces
illegal instruction messages.
gcc 8.3, 9.2 are the latest working gcc version. git bisect between
gcc 8.3 and 8.4 allowed to identify the commit that introcuced the
regression.
Reverting this patch allowed to produce a working rootfs.
Fixes:
https://gitlab.com/buildroot.org/buildroot/-/jobs/786589934
Signed-off-by: Romain Naour <romain.naour@gmail.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
