The default x86-64 architecture for target gcc (ie, the one in poky
build appliances) is native. Since we have a variety of build systems
it will occasionally produce instructions that don't work on all of
our development system.
Instead, set gcc's default architecture to the one specified in
TUNE_CC_ARCH, that guarantees that gcc-runtime and any binaries
produced are compatible with the target machine type.
(From OE-Core rev: 52b952e474e655f8b4b6501813d57e20c9f02ba2)
Signed-off-by: Daniel McGregor <daniel.mcgregor@vecima.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Fix multilib sdk issue where gcc is unable to find linker. Previous
fix was in cdd86896c8d29135f937968e9aa07f919cf543d3 using real-ld
symlink, but that prevented switching between bfd and gold linkers.
Running compiler with debug arguments shows that collect2 tries and
fails to find linker using the multilib triples:
$ $CC -v -Wl,-debug
...
Looking for 'real-ld'
Looking for 'collect-ld'
Looking for 'mips-oemllib32-linux-ld'
Looking for 'mips-oe-linux-mips-oemllib32-linux-ld'
...
collect2 version 12.2.0
ld_file_name = not found
...
collect2: fatal error: cannot find ‘ld’
Using --with-plugin-ld=ld in gcc-cross-canadian builds to set default
linker name for collect2, lets it find the linker correctly:
Looking for 'real-ld'
Looking for 'collect-ld'
Looking for 'ld'
...
collect2 version 12.2.0
ld_file_name = /usr/local/oecore-x86_64/sysroots/x86_64-oesdk-linux/usr/libexec/mips-oe-linux/gcc/mips-oe-linux/12.2.0/ld
Swith between bfd and gold linker works as expected now:
$ $CC -v -Wl,-debug -fuse-ld=gold
...
Looking for 'real-ld'
Looking for 'collect-ld'
Looking for 'ld.gold'
...
collect2 version 12.2.0
ld_file_name = /usr/local/oecore-x86_64/sysroots/x86_64-oesdk-linux/usr/libexec/i686-oe-linux/gcc/i686-oe-linux/12.2.0/ld.gold
(From OE-Core rev: cf1bb16b7f9f81514feaf1e4ecffd9039387bb89)
Signed-off-by: Samuli Piippo <samuli.piippo@qt.io>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This reverts commit cdd86896c8d29135f937968e9aa07f919cf543d3.
real-ld is always used if that is found, which means you cannot
switch between bfd and gold linkers using -fuse-ld gcc option.
(From OE-Core rev: 53ae417c63a4a7ff4a729d3653a31cf1c0758f10)
Signed-off-by: Samuli Piippo <samuli.piippo@qt.io>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This patch appears to be obsoleted by 217d5be, in GCC 11.1 onwards.
libstdc++: Test errno macros directly for all targets [PR 93151]
This applies the same changes to the djgpp and mingw versions of
error_constants.h as r11-6137 did for the generic version.
(From OE-Core rev: e3157da1408b4e6281fdf28825b79c266e93c05e)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This was fixed upstream in de6f402a, as part of 12.2.
(From OE-Core rev: d5e6768f30579abc8559aab60429a292a4c49f3d)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This patch is no longer needed, the upstream Makefile exports CPP from
CPP_FOR_BUILD since 12.1.
(From OE-Core rev: c7546939574dc5e976caae8f0e705b805962ff03)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The -Werror=poison-system-directories argument was being passed to runs
where _just_ CPP was being invoked, so it wasn't passed to runs which
also compile. Add the options to cc1_options so that compile runs also
have fatal poisoning errors.
(From OE-Core rev: e903b29f89e81de244ac77da464c71b718c6854d)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This code wasn't changing the linux.h on i686. Make the entry match that
for i586 which was correct.
This fixes problems where the wrong dynamic loader path was used by
nativesdk-gcc on i686 SDK targets by ensuring SYSTEMLIBS_DIR is replaced
by %r in the correct header files.
(From OE-Core rev: d30eb681f41bf9e921f7f0d42747fff7a4be9229)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Tweak the powerpc code to just include filenames rather than full paths
to avoid build reproducibility issues.
(From OE-Core rev: a71832a14b6944da10c1916861545577f2f2de2c)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add a patch to gcc so that relative paths are handled by -fdebug-prefix-map
and friends. In OE we use relative paths in autotools and removing that
creates a lot of issues we'd have to fix. This alternative allows us to
fix the paths within gcc and improve our debug file coverage (and SPDX
manifests) accordingly.
(From OE-Core rev: 591cbe7140f01470b5799bc51e5b23f58bf95ffa)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Now that we're using absolute paths to run configure, there are absolute
path symlinks within gcc's output. Use our script that fixes these so
that the sstate objects work correctly.
(From OE-Core rev: 78f0bb21ae4337fdf628b2436c6182c32ed4fa9d)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We don't really need to keep S and B separate for debug source purposes
and there shouldn't be source references in WORKDIR that isn't S and B
either.
Separating these out simplifies the shared-work directory handling for
gcc and should also help fix external source usage. Therefore handle
S and B in DEBUG_PREFIX_MAP separately and clean up other code.
Indentation is reduced here as it is introduced on every compiler
commandline so minimising it is helpful.
(From OE-Core rev: c39b5020b8705d17e3745c41e38d0f99a1ac94cf)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This code was some of the earliest reproducible build work we did. To
correctly handle the encoding of file paths, we used relative build
paths to run configure which resulted in relative build paths in the
binaries.
We now have more modern approaches used elsewhere with the prefix remapping
options. These work best with absolute paths, not relative ones. As such,
drop the relative path mangling and switch to using prefix mapping
exclusively on absolute paths.
This makes the code matc the rest of the system and triggers the correct
code to be added in /usr/src/debug.
We have to include both file-prefix and debug-prefix since the assembler
only looks at debug-prefix.
(From OE-Core rev: 309e5d8bb56eb8599d756831f0bd38e6a50cfb05)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ignore the rmdir cmd if using usrmerge distro feature since the
intention is to delete /lib or /lib64 but not libdir under /usr and
base_libdir = libdir when usrmerge is enabled in distro
(From OE-Core rev: 3b7f6b0e0f1f16f89cd924aac001b4f661c145ca)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
some standalone targets e.g. riscv64-elf disable shared linking for
baremetal ELF ABI in ld, therefore lets make it a static library
(From OE-Core rev: 3c6219dfcbcbde314648ba8cc54a90b32ea1c952)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This disables tests for runtime system specific functions like fcntl
there is no other change seen
+configure:77372: checking for fcntl
+configure:77372: result: yes
In libsttdc++ configure log outputs.
[YOCTO #14803]
(From OE-Core rev: 586b6239252ecf2711930b60f7f0f617e38ac737)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This is a dummy shared object therefore reduce dependencies further by
not requiring the C startup files, we wont use this shared library for
anything useful anyway
(From OE-Core rev: e1c4953f5587621d4911416c2e9350ad0c27b296)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This fixes reproducibility issues with multilibs were a different recipe
specific sysroot is used which was leaking into debug symbols in libraries.
(From OE-Core rev: f442edf51e256bd315bd8e4ac4d9fa12b8e9e092)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If gold is enabled as the default linker, it errors trying to link
to our dummy library empty file and this turns off things which should
be present in libstdc++.
For example, _GLIBCXX_HAVE_S_ISREG isn't defined and HAVE_S_ISREG in
libstdc++-v3/config.h isn't set properly.
Instead of just creating an empty file, create an empty elf binary
instead which addresses the issue.
(From OE-Core rev: 2070bcd10aa3a05c96c8501c6a8c1e129fb1d440)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
There is a pattern that several recipes need to break the dependency of ${PN}-dev
on ${PN}, most often as ${PN} may be be empty. Add a new variable to parameterise
this and allow it to be changed more easily.
(From OE-Core rev: a5b381c0f45c590a762647a9956a8f41e2e2315e)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Some tasks may reference ${B} for gcc-source which in general would not exist.
It has dependencies on HOST_SYS and TARGET_SYS which are not appropriate for a
shared recipe like gcc-source. This causes problems for the archiver and
multiconfigs in particlar.
Set B to something else to avoid these task hash issues.
Acked-by: Jose Quaresma <jose.quaresma@foundries.io>
(From OE-Core rev: beb2a76c591e985c6fc7ed473abd1bee27f955a2)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
To match the changes to gcc-cross, add a nativesdk-zstd dependency to ensure
our configurations match.
(From OE-Core rev: c145e00710fe557b5a3832fdc556ad53434b3615)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Sharing sstate cache binaries between two systems, one with libzstd installed
and the other without, leads to various gcc components being linked against
the system libzstd and failing to run on the system with out it installed.
Make zstd-native from our system available.
(From OE-Core rev: 2b0a6f03137f24b211c8881cebf65732e550a942)
Signed-off-by: Jeremy A. Puhlman <jpuhlman@mvista.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The patch is needed in order to support recent glibc (2.34).
libsanitizer/ChangeLog:
PR sanitizer/101749
* sanitizer_common/sanitizer_posix_libcdep.cpp: Prevent
generation of dependency on _cxa_guard for static
initialization.
(From OE-Core rev: c44c4e7fb3c860d9fcb2aada0c9d4acb1e1e8101)
Signed-off-by: Sundeep KOKKONDA <sundeep.kokkonda@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
In commit ceda3238 (meta/meta-selftest/meta-skeleton: Update LICENSE
variable to use SPDX license identifiers) all LICENSE variables were
updated to only use SPDX license identifiers.
This does the same for comments and other variables where it is
appropriate to use the official SPDX license identifiers. There are
still references to, e.g., "GPLv3", but they are then typically in
descriptive text where they refer to the license in a generic sense.
(From OE-Core rev: 165759dced7fbe73b1db2ede67047896071dc6d0)
Signed-off-by: Peter Kjellerstedt <peter.kjellerstedt@axis.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
An automated conversion using scripts/contrib/convert-spdx-licenses.py to
convert to use the standard SPDX license identifiers. Two recipes in meta-selftest
were not converted as they're that way specifically for testing. A change in
linux-firmware was also skipped and may need a more manual tweak.
(From OE-Core rev: ceda3238cdbf1beb216ae9ddb242470d5dfc25e0)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Move cc1plus from gcc to g++ package. Therefor, remove the duplicate
FILES entry from gcc package and keep the entry in g++ package.
(From OE-Core rev: f16c0efdeaca7cd08bd5609141b1094139e746f9)
Signed-off-by: Stefan Herbrechtsmeier <stefan.herbrechtsmeier@weidmueller.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The glob to remove the gcc-<version> binary expects a single-digit major
version which is no longer true.
(From OE-Core rev: 30de10cf939cf65c0cab6730a3575bced5c7533a)
Signed-off-by: Stefan Herbrechtsmeier <stefan.herbrechtsmeier@weidmueller.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The patch to gcc to disable use of libstdc++ when configuring the library
during gcc-runtime is old and there are perhaps better ways to do this now.
If removed, most builds still "work" but incorrect values for things like
atomic ops are found during configure. mips64 and ppc fail with on target
simple tests of g++ in testimage.
Instead we can create a dummy libstdc++ which allows the correct configure
test results to be obtained.
Discussed with upstream about the patch are ongoing, if accepted, we can
switch back to the commandline option if it is added in future.
(From OE-Core rev: 34b0edb0d3120c32063ff7e3dd52be20d60401d5)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
While building GCC it checks whether the include directory exists,
if it doesnt it throws an error and exits:
| The directory that should contain system headers does not exist:
| tmp/work/x86_64-nativesdk-pokysdk-linux/gcc-cross-canadian-riscv32/
11.2.0-r0/recipe-sysroot/usr/include/
| make[1]: *** [Makefile:3257: stmp-fixinc] Error 1
Even though for the baremetal toolchain not having this directory
does make sense.
We overcame this by removing the --with-sysroot=/not/exist argument
for baremetal toolchains (via TARGET_OS override), however, the
newlib toolchain does have headers and an includedir, hence by fixing
the baremetal toolchain we broke the newlib one since it uses the same
TARGET_OS as baremetal, causing for example (on newlib):
/sdk/sysroots/x86_64-pokysdk-linux/usr/lib/riscv32-poky-elf/gcc/
riscv32-poky-elf/11.2.0/include/stdint.h:9:16:
fatal error: stdint.h: No such file or directory
| # include_next <stdint.h>
^~~~~~~~~~
compilation terminated.
By creating a dummy includedir, and removing the previous fix we
allow GCC to be built the same way, unifying the cross compiler
for all targets.
After this fix both TCLIBC=baremetal and TCLIBC=newlib SDKs work
properly.
(From OE-Core rev: 08ff7c87efdf373f1874fcccd9a2a73fc0efef30)
Signed-off-by: Alejandro Enedino Hernandez Samaniego <alejandro@enedino.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The "gcc: Fix compile of gcc plugins" patch had a typo making it a noop.
This was due to a rework to make it specific to the linux64.h header.
(From OE-Core rev: 5f0f00f0988c02d7fe0fa62edf3e4901ce9941a6)
Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Without this, compiling gcc plugins will fail with an error similar to:
[...]
fatal error: ./config/i386/linux64.h: No such file or directory
[...]
In Yocto, we set up compiling gcc-cross out of tree. Which in turn makes
the generated headers end up in B. The tm.h header will include
generated headers that are expected in plugin/include/config/*.
For example, the linux64.h header, when generating gcc-cross for x86-64,
will end up in tm.h header as:
include "./config/i386/linux64.h"
On the other hand, the make rule `install-plugin` in gcc/Makefile.in
will install the linux64.h assuming that it is generated in the sources
directory and because this is not the case in our setup, the Makefile
ends up installing it in plugin_includedir/`basename $$path` which ends
up installing the header in [..]plugin/include as opposed to
[..]plugin/include/config/i386 (as expected by the generator of tm.h).
The included patch modifies the Makefile rule to match the assumption of
gcc-cross being compiled out of tree.
(From OE-Core rev: 92167f8e02bb6fbbe1ee6a6678525a0ae27b00a5)
Signed-off-by: Andrei Gherzan <andrei.gherzan@huawei.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This patch has been accepted upstream
(From OE-Core rev: 4c644708798f258d45dbcfa01909626de32710d4)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The CVE applies to binutils 2.26 and not to gcc so ignore there.
(From OE-Core rev: fea2726663a3db03170c49fceaffc632c509aeea)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
According to comments on the bug report from gcc developers, we
no longer need to do this post gcc 10. Lets therefore drop the patch.
(From OE-Core rev: 8ddc6f46d40cdcc462de23d1b1218f2ed9fd3d07)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
gcc-configure-common.inc already sets --with-abi=64 for our mips64
targets so this patch is no longer needed.
[YOCTO #14639]
(From OE-Core rev: e0010af733825ed1050fd3342cf3ef1c478df1a0)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This patch was mentioned upstream a long time ago:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=47256
Changes from gcc 10 onward mean it is no longer needed as mentioned in the
above bug report. Drop the patch.
(From OE-Core rev: 6fca075a8d87d033359f81d476c8506df66ad53d)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
