- New argument android_exe_type for executables
- Changes to the b_sanitize option
- New C standard c2y (and gnu2y)
- i18n module xgettext
- version_compare now accept multiple compare strings
- Improvements to Objective-C and Objective-C++
- Per project subproject options rewrite
- objects added correctly to Rust executables
- rust.test now supports link_whole
- Meson can run "rustdoc" on Rust projects
- The Wayland module is stable
- New swift_std compiler option
- New option to execute a slice of tests
- Valgrind now fails tests if errors are found
Full release notes: https://mesonbuild.com/Release-notes-for-1-8-0.html.
(From OE-Core rev: 84ded53ea7e4db1e3d7ea60527374040ec64b4fd)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
libbfd is enabled and it invokes libtool relink step to get rid of
rpaths during install step. libtool 2.4 patches rightly prepend
lt_sysroot to target libdir which relink step would use, such a patch
is also upstreamed into libtool [1]
However, when we do not specify --with-libtool-sysroot, lt_sysoot prefix
turns out to be empty and relink step adds -L/usr/lib to linker command
line. it cause all sort of problems e.g. when the build host has zlib
installed then libbfd tries to link with /usr/lib/libz.so instead of the
one in target sysroot, fortunately OE linker reports host path contamination
and fails the build. Second problem seen when building target binutils on
musl, it links with libc.so correctly but libtool relink step resolves
libc.so in /usr/lib due to this path and expands it into libc.so.6 and
ld-linux-x86-64.so.2 ( glibc host's libc.so is a linker stub ) in
DT_NEEDED ELF section which does not work with
musl linker and build QA detects this anamoly as well.
ERROR: binutils-2.44-r0 do_package_qa: QA Issue: /usr/lib/libbfd-2.44.so contained in package libbfd requires libc.so.6(GLIBC_2.34)(64bit), but no providers found in RDEPENDS:libbfd? [file-rdeps]
This inconsistency may cause produced binaries mistakenly linked against
libraries from the host distro.
Add libtool sysroot to point libtool to correct sysroot location during
build.
[1] https://git.savannah.gnu.org/cgit/libtool.git/commit/?id=3221f9f0fb98d5740ab5d0e8db6a731302520644
(From OE-Core rev: 89e984bd7edb9a2013b65d0215fd6af9aadd59e8)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We have long since obsoleted md5sum in favour of sha256sum. Drop the remaining
56 entries (which were showing many recipes hadn't been touched in a long time).
They all do have the corresponding sha256sum entries as is clear from the diff.
(From OE-Core rev: 7e4bfcc9706fa8a09f6a0004174a2c3b21c90df3)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
I previously upgrade autoconf 2.72e (the fifth release candidate) to the
final release 2.72, but unfortunately failed to notice that the version
appears to go backwards as the package managers sort 2.72e _after_ 2.72.
We could resolve this with an epoch (via PE) but epochs are forever and
this mistake is temporary. We could set a convoluted PV but that gets
messy as it then means SRC_URI needs to set manually. Instead, set a
convoluted PKGV to fix the sorting order in a way that only alters the
output packages and nothing else, which can be removed when the next
major release is made.
[1] oe-core 11ff8dba0ce ("autoconf: upgrade to final 2.72 release")
(From OE-Core rev: 58b185b25dff627eecc980919dd44dcce32fbca2)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Fix the numpy ptests by doing the following:
- Add meson to ptest RDEPENDS in the recipe;
- Add python3-unittest-automake-output as a ptest RDEPENDS;
- Convert run-ptest to a shell script that sets PYTEST_DEBUG_TEMPROOT to
a directory inside the same path that contains the script, create that
directory, and then invoke the tests with `pytest --automake` pointed
at the numpy path in site-packages (copying the tests into the normal
PTEST_DIRECTORY seems to cause module import breakages in some tests).
This also includes skipping two problematic tests which require a C
compiler and use up a lot of space, respectively;
- Set 'IMAGE_ROOTFS_EXTRA_SPACE = "3048576"' for python3-numpy in
core-image-ptest.bb;
- Also set 'QB_MEM:virtclass-mcextend-python3-numpy = "-m 4096"' in
core-image-ptest.bb;
- Move python3-numpy from the PTESTS_PROBLEMS list to the PTESTS_SLOW
one.
Results on qemux86-64:
Testsuite summary
DURATION: 87
END: /usr/lib/python3-numpy/ptest
2025-04-29T17:35
STOP: ptest-runner
TOTAL: 1 FAIL: 0
Note that many of the skipped tests are due to the absence of
python3-mypy (which is currently in meta-python) and of a Fortran
compiler.
(From OE-Core rev: 7f0b5e8faa1b246531ac425c99a629eb344b21e2)
Signed-off-by: Trevor Gamblin <tgamblin@baylibre.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Below commits on binutils-2.44 stable branch are updated.
fe459e33c67 PR 32603, revert message changes in 0b7f992b78fe and 31e9e2e8d109
37d12dd25d8 gdb/compile: add missing entry in bfd_link_callbacks array
31e9e2e8d10 PR 32603, more ld -w misbehaviour
0b7f992b78f PR 32603, ld -w misbehaviour
8cb98edf123 s390: Add support for z17 as CPU name
ed70d86b491 x86: Remove AVX10.2 256 bit rounding support
e1af7e590a5 elf: Clear the SEC_ALLOC bit for NOLOAD note sections
35db8c6dd2f ld: Pass -Wl,-z,lazy to compiler for i386 lazy binding tests
cc7ec316a45 Updated translations for bfd and gold
bf088ee09a7 PR 32731 ub sanitizer accessing filenames_reversed
78082591ec7 score-elf gas SEGV
d4c7ee9fbc1 gas: fix rs_fill_nop listing
a68d096a0ab Open the 2.44 branch for further development
Dropped: 0015-CVE-2025-1153.patch
Testing was done and there were no regressions found
(From OE-Core rev: 749bdd5bda6f9b5af8d63794858bbb9834294fcb)
Signed-off-by: Deepesh Varatharajan <Deepesh.Varatharajan@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Even with our fixes in deterministic_imports.patch the
importlib.metadata package scan was still returning Distribution objects
for empty directories. This interacts badly with rebuilds when recipes
are changing as when a recipe is removed from the sysroot directories
are not removed[1].
In particular this breaks python3-meson-python-native rebuilds when
Meson upgrades from 1.7 to 1.8: the site-packages directory has an empty
meson-1.7.dist-info/ and populated meson-1.8.dist-info/. Whilst it's
deterministic to return the empty 1.7 first, this breaks pypa/build as
it looks through the distributions in order.
We had discussed this with upstream previously and there's a more
comprehensive fix upstream (actually in importlib_metadata, not cpython)
which ensures that valid distribution objects are listed first. So we
can drop our patch and replace it with a backport to fix these rebuilds.
[1] oe-core 4f94d929639 ("sstate/staging: Handle directory creation race issue")
(From OE-Core rev: 73de8daa6293403f5b92d313af32882c47bce396)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The user cannot override the setup command options --cross-file and
--native-file because the meson-wrapper places these options on the
command line after the user options. This problem was noticed when
trying to build with an SDK using a custom cross-file.
(From OE-Core rev: 045a53349a5c00b318feb7029470d3fb084b61c8)
Signed-off-by: Tom Hochstein <tom.hochstein@oss.nxp.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Due to commit [cdrtools-native: fix booting EFI ISO live failed]
applied to improve mkisofs to fix nsectors exceeds 0xffff situation
which set selection criteria type = 2 and save extra nsectors to
vendor unique selection criteria
In following case, add 64MB extra space to bootable image efi.img,
and the partition table of EFI is truncated to 32M
$ echo 'IMAGE_FSTYPES:pn-core-image-minimal = " live"' >> conf/local.conf
$ echo 'MACHINE_FEATURES:append = " efi pcbios"' >> conf/local.conf
$ echo '# 64MB extra space to bootable image efi.img' >> conf/local.conf
$ echo 'BOOTIMG_EXTRA_SPACE = "65535"' >> conf/local.conf
$ bitbake core-image-minimal
$ fdisk -l tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso
...
Device Boot Start End Sectors Size Id Type
tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso1 * 0 376831 376832 184M 0 Empty
tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso2 120 65654 65535 32M ef EFI (FAT-12/16/32)
After applying this patch to process extra sector count, the partition
table of EFI is 90.3M
$ fdisk -l tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso
...
Device Boot Start End Sectors Size Id Type
tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso1 * 0 376831 376832 184M 0 Empty
tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso2 120 185151 185032 90.3M ef EFI (FAT-12/16/32)
[1]https://pdos.csail.mit.edu/6.828/2017/readings/boot-cdrom.pdf
(From OE-Core rev: b4e112ed7e6ba5a6c6df530d696485a588831851)
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
In ISO live, if the size of efi.img > 32MB, and copy EFI application
(bootx64.efi) to efi.img behind of kernel and initrd, UEFI system
could not find EFI application bootx64.efi
Using QEMU+OVMF to boot ISO live image, press ESC to enter UEFI shell:
...
Shell> ls FS0:\
Directory of: FS0:\
04/05/2011 23:00 12,985,344 bzImage
04/05/2011 23:00 <DIR> 2,048 EFI
04/05/2011 23:00 20,494,696 initrd
04/05/2011 23:00 26 startup.nsh
3 File(s) 33,480,066 bytes
1 Dir(s)
Shell> ls FS0:\EFI
Directory of: FS0:\EFI
0 File(s) 0 bytes
0 Dir(s)
...
In following case, add 64MB extra space to bootable image efi.img,
and the partition table of EFI is truncated to 26.3M
$ echo 'IMAGE_FSTYPES:pn-core-image-minimal = " live"' >> conf/local.conf
$ echo 'MACHINE_FEATURES:append = " efi pcbios"' >> conf/local.conf
$ echo '# 64MB extra space to bootable image efi.img' >> conf/local.conf
$ echo 'BOOTIMG_EXTRA_SPACE = "65535"' >> conf/local.conf
$ bitbake core-image-minimal
$ fdisk -l tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso
...
Device Boot Start End Sectors Size Id Type
tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso1 * 0 376831 376832 184M 0 Empty
tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso2 120 54079 53960 26.3M ef EFI (FAT-12/16/32)
According to page 11: `Figure 5 - Section Entry' in El Torito Bootable
CD-ROM Format Specification [1]. The sector count takes 2 byte which
means max sector count is 0xffff (65535), for 512-byte sector, the
size of bootable image is no more than 32MB (65536 * 512 / 1024 / 1024)
This commit truncate to 32MB if image size larger than 32MB, and
report a warning, then save the extra image sector count to
vendor unique selection criteria
After apply this commit, the partition table of EFI is truncated to 32M
$ fdisk -l tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso
...
Device Boot Start End Sectors Size Id Type
tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso1 * 0 376831 376832 184M 0 Empty
tmp/deploy/images/qemux86-64/core-image-minimal-qemux86-64.rootfs.iso2 120 65654 65535 32M ef EFI (FAT-12/16/32)
[1]https://pdos.csail.mit.edu/6.828/2017/readings/boot-cdrom.pdf
(From OE-Core rev: 259bb8907d8bfe1217e88a3b6935c160e5a92f8d)
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
* on hosts with gcc-15 or whenever glib PACKAGECONFIG isn't enabled
and pkgconfig uses own old bundled glib
* fixes:
http://errors.yoctoproject.org/Errors/Details/853015/
../../../git/glib/glib/goption.c:169:14: error: two or more data types in declaration specifiers
169 | gboolean bool;
| ^~~~
../../../git/glib/glib/goption.c:169:18: warning: declaration does not declare anything
169 | gboolean bool;
| ^
(From OE-Core rev: aada741d57e181ff35e0715012513dc9d669eaef)
Signed-off-by: Martin Jansa <martin.jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The current dpkg runtime version is generated by 'git describe' command
in build-aux/get-version. But since we have made changes to git repo
locally, it adds a -dirty suffix to the version number.
Add a .dist-version file in the source directory. Then get-version can
determine the version through .dist-version instead of "git describe"
command, thus removing the -dirty suffix.
Before the fix:
$ dpkg --version
Debian 'dpkg' package management program version 1.22.11-dirty (amd64).
After the fix:
$ dpkg --version
Debian 'dpkg' package management program version 1.22.11 (amd64).
(From OE-Core rev: 04d00d69941a9d09bfbb0956186db7a7d0961327)
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
To adapt user network enviroment, buildtools should first try to use
the user configured envs like SSL_CERT_FILE/CURL_CA_BUNDLE/..., if these
envs is not set, then use the auto-detected ca file and ca path, and
finally use the CA certificates in buildtools.
nativesdk-openssl set OPENSSLDIR as "/not/builtin", need set SSL_CERT_FILE/SSL_CERT_DIR to work
nativesdk-curl don't set default ca file, need
SSL_CERT_FILE/SSL_CERT_DIR or CURL_CA_BUNDLE/CURL_CA_PATH to work
nativesdk-git actually use libcurl, and GIT_SSL_CAPATH/GIT_SSL_CAINFO
also works
nativesdk-python3-requests will use cacert.pem under python module certifi by
default, need to set REQUESTS_CA_BUNDLE
(From OE-Core rev: 8a7ec52e9b35654bee48cd948c6c34c63db3e265)
Signed-off-by: Changqing Li <changqing.li@windriver.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
rpm-sequoia-crypto-policy tries to validate the configuration files
using host tools. For the Strongswan policy, it uses
"ipsec readwriteconf" which is not available on Debian 12 with
Strongswan installed.
To fix this, add and use an option to skip the problematic validation.
(From OE-Core rev: d10ca0fe194b62b2f383be880a008cde2bd0fd4f)
Signed-off-by: Yoann Congal <yoann.congal@smile.fr>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Some unit tests were previously marked as ignored due to failures in earlier
versions of Rust. With the upgrade to Rust 1.85.1, these tests are now passing
consistently. They've been re-enabled and verified to run successfully on the
latest version.
(From OE-Core rev: 65c207fb034848f9ef0de8cf8725c5eded363aca)
Signed-off-by: Deepesh Varatharajan <Deepesh.Varatharajan@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This allows to enable sdl-image support from a bbappend file
(From OE-Core rev: ca848cd15bf61f12cdadf4c94c985f2e87a932c8)
Signed-off-by: Markus Volk <f_l_k@t-online.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
-ffile-prefix map is more comprehensive when it comes to reproducible
builds and its superset of all prefix-mapping options in compilers
This makes is cleaner and workable across gcc and clang, clang does not
support -fcanon-prefix-map and it has to be explicitly omitted when using
clang.
There are lambdas generated in templates by clang which still get the
absolute paths despite -fdebug-prefix-map, this helps with that as well.
nasm is an outlier and we have fixed it by adding -fdebug-prefix-map option
luckily we do not pass DEBUG_PREFIX_MAP to nasm, in all recipes which use
nasm either pass -fdebug-prefix-map explicitly to nasm or they rewrite it
to use nasm flags syntax.
We have discussed this in past [1]
[1] https://patchwork.yoctoproject.org/project/oe-core/patch/20230428032030.2047920-1-raj.khem@gmail.com/#10281
(From OE-Core rev: ff73fa7ef7666a6dbe34f15515bc3ab6e574c5b0)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Cc: Jacob Kroon <jacob.kroon@gmail.com>
Cc: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
autoconf 2.72e was the fifth release candidate build, and whilst we had
good reason to use RCs during development we failed to notice that the
final release was made because alpha.gnu.org _only_ contains the release
candidates.
There are minimal changes in the final release compared to 2.72e:
6199cff2 (tag: v2.72) Finalize NEWS for release 2.72.
This is the finalisation of the release, updating documentation and
version numbers.
e544a323 Save/restore AC_LANG setting in AC_PROG_CC_C_O.
This is a bug fix, aligning behaviour of AC_PROG_CC_C_O.
634fb3e7 Remove bashism from tests/local.at.
4fbe6436 Revert debugging change committed by mistake
8794cd1c Address “autom4te preselections” testsuite failure
e5d98f4c Avoid a spurious test failure with shells that ignore write errors.
3fbfb13e Address “autoconf: forbidden tokens, basic” testsuite failure
fdb58df6 testsuite: Skip tests that use aclocal or automake if they’re too old.
These are fixes for the test suite.
(From OE-Core rev: 11ff8dba0cee801e93cf4bdb3ccf043f4d2f751f)
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Backport patch to update __version__ to 0.2 to match current version.
(From OE-Core rev: 67332a80391cb341c57f95e8c7b6bab306e25b1c)
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Some unit tests were previously marked as ignored due to failures in earlier
versions of Rust. With the upgrade to Rust 1.85.1, these tests are now passing
consistently. They've been re-enabled and verified to run successfully on the
latest version.
(From OE-Core rev: 775108f131412122beb2b844dc5e2a4d47ce3b89)
Signed-off-by: Deepesh Varatharajan <Deepesh.Varatharajan@windriver.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
>From https://metacpan.org/pod/perldelta :
This provides a fix for CVE-2024-56406 - Heap buffer overflow vulnerability with tr//
There are no changes intentionally incompatible with 5.40.1
(From OE-Core rev: 53c772c941d5ce69ab89f35cd04724b740daf269)
Signed-off-by: Jason Schonberg <schonm@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Some projects are picky and want to hanve llvm-min-tblgen in addition to
a normal llvm-tblgen. Install the tool to the sysroot.
(From OE-Core rev: b4acbadc16733c29024694ba0e5c0987dc7b459c)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Import a fix from meta-clang, which allows packages using LLVM to find
tools from the native sysroot.
(From OE-Core rev: 112b9d4c4c10d7ac0e2a8bc239d451b84b90c82c)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
There is a libc patch needed to fix rustix 0.38.x build
It is already fixed in latest libc upstream, backport it
here
(From OE-Core rev: 7fd3871f21ed9039e0b3b27840109b460b6991ac)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
>From gdb documentation:
> There are two ways that highlighting can be done. First, if
> gdb was linked with the GNU Source Highlight library, then it
> is used. Otherwise, if gdb was configured with Python
> scripting support, and if the Python Pygments package is available,
> then it will be used.
(From OE-Core rev: b1106ed570a237f3ab80ba3fedb086c9ec0cb4a8)
Signed-off-by: Etienne Cordonnier <ecordonnier@snap.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Here is one testcase:
For recipe tensorflow-lite-host-tools_2.18.0.bb, refer [1],
do_configure[network] = "1"
and it will git clone some repos in CMakeLists.txt
When buildtools is used and nativesdk-git is installed into sdk,
do_configure failed with error:
[1/9] Performing download step (git clone) for 'protobuf-populate'
Cloning into 'protobuf'...
fatal: unable to access 'https://github.com/protocolbuffers/protobuf/': error setting certificate file: /usr/local/oe-sdk-hardcoded-buildpath/sysroots/x86_64-wrlinuxsdk-linux/etc/ssl/certs/ca-certificates.crt
Fix by adding GIT_SSL_CAINFO in BB_ENV_PASSTHROUGH_ADDITIONS, so that
user can export GIT_SSL_CAINFO=${GIT_SSL_CAINFO} in their
do_configure:prepend() to fix above do_configure failure
CURL_CA_BUNDLE and REQUESTS_CA_BUNDLE is similar envvars, so all add
into BB_ENV_PASSTHROUGH_ADDITIONS
[1] https://github.com/nxp-imx/meta-imx/blob/styhead-6.12.3-1.0.0/meta-imx-ml/recipes-libraries/tensorflow-lite/tensorflow-lite-host-tools_2.18.0.bb
(From OE-Core rev: 183e043de423fd3f7833366ca524a6f7d17e6d14)
Signed-off-by: Changqing Li <changqing.li@windriver.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
* make git,curl,python3-requests align with openssl, move the setting of
envvars into respective envfile
* for environment.d-openssl.sh, also check if ca-certificates.crt exist
before export envvars
(From OE-Core rev: 9d8ebfcd55ad274e79cb81f162397c8f39f4287b)
Signed-off-by: Changqing Li <changqing.li@windriver.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Import bindgen-cli from the meta-clang layer, as it is required to
generate bindings for Mesa / RustiCL. Specify Khem as maintainer as he
is maintaining the meta-clang layer, which included this recipe too.
(From OE-Core rev: 8cd697f68eeaa7c19f97a0f44d0c5c0cc3252040)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Pull in the list of targets being enabled by meta-clang in order to
make the libllvm from OE-Core runtime compatible with the one built by
the clang recipe inside meta-clang.
(From OE-Core rev: 672a21e16a2f4aadc588b367f0fd1ff1df7b5c67)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Mesa builds require SPIRV-LLVM-Translator in order to build CLC support.
While it is possible to build it separately (like meta-clang does), it
also requires a separate clang compiler (which OE-Core doesn't provide).
Use an alternative path and build SPIRV-LLVM-Translator as a part of the
LLVM build process, lifting the requirement for the separate clang. Once
Clang is properly integrated into the OE-Core layer, this can be split
to a separate package and dropped. In order to ease migration from
meta-clang and to ease future split, PROVIDE the spirv-llvm-translator
and make the llvm-spirv package RPROVIDE spirv-llvm-translator.
(From OE-Core rev: a617aaf3ac6b1ef2f4b0aaf6ca60122425ea6a7e)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Current Mesa requires libclc in order to build Intel Gallium (iris) or
Vulkan drivers. In order to unblock Mesa updates, build libclc as a part
of the LLVM build process.
The libclc requires Clang for building, so enable Clang too via
LLVM_PROJECTS. Remove clang from the DESTDIR, so that it is only used
during the build process. Proper handling of Clang as a toolchain is a
separate and much more complicated topic. Clang is enabled for all kinds
of builds as it is a requirement for libclc to build.
Cleaning of ${datadir} has been moved to happen at the end of do_install
to be performed after do clang removal and made not to fail if
${datadir} is not empty, as libclc install files into that location.
NVPTX and SPIRV targets are enabled since they are required for the
libclc to link.
LLVM_NATIVE_TOOL_DIR is used to pass a path to the native tools built
via llvm-native instead of building them during the target and/or
nativesdk build.
The llvm-clc package is set to RPROVIDE libclc (and the recipe also
PROVIDEs libclc) in order to be compatible with the meta-clang packages
(and to be future-compatible with the split package after proper Clang
support is introduced to OE-Core).
(From OE-Core rev: 2773404dec7840e2252cfd5bf474be70d8e7888f)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
On native OE builds the LLVM_LDFLAGS gets a lot of extra flags which may
depend on exact build dir and as such be inappropriate to binaries
shared through sstate-cache. It is not possible to override this
behaviour in llvm-config since crossscripts are not being used for
native builds. Ignore LLVM_LDFLAGS for native builds.
(From OE-Core rev: a93d2bdf5c0cf1c2c08d99fd4ccd8d1aae0e0ba5)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The recipes targeting the SDK (e.g. nativesdk-mesa) should also use the
llvm-config wrapper, otherwise they get get a lot of incorrect flags,
the most offensive being the rpath, as it results in TMPDIR being a part
of the resulting binary.
(From OE-Core rev: 4d4b02d28c8c5807c2a90402c1239ed2581eb884)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The llvm-config wrapper also needs to override a --libdir switch,
otherwise software (like mesa) end up using TMPDIR-based directory for
Clang resource lookup (and also encoding TMPDIR into the target binary).
(From OE-Core rev: c804ef9516d840e9bef717b6add77783a21a443a)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Instead of pushing an `llvm` sourcedir into S, specify it through the
OECMAKE_SOURCEPATH so that all the patches are applied to the topdir and
debug paths also use top path to map.
(From OE-Core rev: 43019c7b303da865a563fd672a699175b3d11e5a)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Neither MAJOR_VER nor MINOR_VER are defined, so libLLVM.so.N.M ends up
in the main libllvm package rather than the llvm-libllvm package, as
expected. Fix this by properly using two-component version.
Fixes: db3bf550a372 ("llvm: Update to 18.1.0 RC4")
(From OE-Core rev: 1f729a019c6e5ea35d9a5196b68d81f31a8a1594)
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
