| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
|
| |
Signed-off-by: Khem Raj <khem.raj@oss.qualcomm.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Remove the riscv toolchain workarounds that are no longer needed with
current clang:
- mesa: stop pinning gcc on riscv64
- frr: stop pinning gcc on riscv64 and riscv32 (llvm-project#61412)
- ffmpeg: stop appending -march=rv64gczbb/rv32gczbb, which also made
the binaries require Zbb on plain rv64gc/rv32gc tunes
- dav1d: stop building with -no-integrated-as on riscv64
AI-Generated: Uses Claude Code
Signed-off-by: Khem Raj <khem.raj@oss.qualcomm.com>
|
| |
|
|
|
|
|
|
| |
The gcc pins were added for a clang-15 crash; drop them so grub and
grub-efi use the default toolchain on riscv64 again.
AI-Generated: Uses Claude Code
Signed-off-by: Khem Raj <khem.raj@oss.qualcomm.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Appending -march=rv64gcv (rv32gcv) to TOOLCHAIN_OPTIONS made every object
in numpy require the vector extension, not just its RVV dispatch targets.
On rv64gc tunes the result crashes with SIGILL as soon as
_multiarray_umath is loaded on a CPU or kernel without V, e.g. a
std::unordered_map constructor in a static initializer is vectorised:
vsetivli zero,2,e64,m1,ta,ma
This breaks the python3-fsspec ptest (via pandas -> numpy) on
qemuriscv64, whose linux-yocto kernel has CONFIG_RISCV_ISA_V unset.
numpy's meson build already compiles the RVV dispatch sources with
-march=rv64gcv on its own (baseline: none, dispatch: RVV), and the
highway __riscv_vsetvl_e8mf8 error this worked around for numpy 2.2.3
no longer happens with numpy 2.5.4 and clang 23.
AI-Generated: Uses Claude Code
Signed-off-by: Khem Raj <khem.raj@oss.qualcomm.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
GI_DATA_ENABLED was forced False for these two on toolchain-clang:x86-64
to work around g-ir-scanner crashing while parsing glibc 2.41 headers
under clang. Verified with clean from-scratch rebuilds of both (clang
23.1.1, glibc 2.44+git, forcing GI_DATA_ENABLED back to True) that
g-ir-scanner/g-ir-compiler now complete cleanly and produce their
.gir/.typelib files with no errors, so the workaround is stale here.
This unblocks sushi, whose html.blp requires the WebKit-6.0 typelib
that webkitgtk only produces when introspection is enabled.
libblockdev, webkitgtk3, and the 32-bit x86 entries are left as-is
since they were not re-verified.
AI-Generated: Uses Claude Code
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Set the layer compatibility to the Yocto Project release Blacksail
following the recent change in openembedded-core:
https://git.openembedded.org/openembedded-core/commit/?id=37623ffbdd34a5c6b3e12ee33195f8e82343f198
Fixes:
ERROR: Layer clang-layer is not compatible with the core layer
which only supports these series: blacksail (layer is compatible
with wrynose)
This work was sponsored by GOVCERT.LU.
Signed-off-by: Leon Anavi <leon.anavi@konsulko.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Replace all_multilib_tune_values with oe.utils.all_multilib_tune_values
following the move functions from utils.bbclass to utils.py in
openembedded-core.
Fixes:
bb.data_smart.ExpansionError: Failure expanding variable
CLANGCROSSCANADIANDEPS, expression was
${@all_multilib_tune_values(d, 'CLANGCROSSCANADIAN')} which
triggered exception NameError: name 'all_multilib_tune_values'
is not defined
Signed-off-by: Leon Anavi <leon.anavi@konsulko.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
perf inherits kernel-arch, which since oe-core 0e3756efe0 ("kernel-arch:
add KERNEL_TOOLCHAIN to switch the whole kernel toolchain") sets
KERNEL_TOOLCHAIN ?= "gcc"
TOOLCHAIN = "${KERNEL_TOOLCHAIN}"
bitbake's basehash records the unexpanded value of a variable along with
its dependency set, not just the final expansion. Overriding TOOLCHAIN
here replaced '${KERNEL_TOOLCHAIN}' with the literal 'gcc' and dropped
the recorded dependency on KERNEL_TOOLCHAIN, so merely adding this layer
changed perf's basehash even though the resulting toolchain is identical.
yocto-check-layer flags that as the layer altering OE-Core signatures:
AssertionError: Adding layer meta-clang changed signatures.
9 signatures changed [...]
List of dependencies for variable TOOLCHAIN changed from
'frozenset({'KERNEL_TOOLCHAIN'})' to 'frozenset()'
Dependency on Variable KERNEL_TOOLCHAIN was removed
Variable TOOLCHAIN value changed from '${KERNEL_TOOLCHAIN}' to 'gcc'
Pin KERNEL_TOOLCHAIN instead. TOOLCHAIN then stays byte-identical to the
OE-Core baseline, so the signatures match, while perf is still forced to
gcc for anyone setting KERNEL_TOOLCHAIN = "clang".
Signed-off-by: Khem Raj <khem.raj@oss.qualcomm.com>
|
| |
|
|
|
|
|
|
|
|
| |
Remove stale dynamic layer references (browser-layer, networking-layer,
meta-python, qt5-layer) from BBFILES_DYNAMIC in layer.conf.
These directories no longer exist under dynamic-layers/, so the
patterns are dead code and have no effect.
Signed-off-by: Deepesh Varatharajan <Deepesh.Varatharajan@windriver.com>
|
| |
|
|
|
|
| |
2025.12 does not compile with clang sadly
Signed-off-by: Khem Raj <khem.raj@oss.qualcomm.com>
|
| |
|
|
|
|
| |
The recipe builds fine with clang now.
Signed-off-by: Ross Burton <ross.burton@arm.com>
|
| |
|
|
|
|
| |
This error doesn't happen anymore, remove the append.
Signed-off-by: Ross Burton <ross.burton@arm.com>
|
| |
|
|
|
|
|
|
| |
Valgrind doesn't have ptests anymore[1] so this append is pointless.
[1] oe-core 8fc31fb235 ("valgrind: disable ptests")
Signed-off-by: Ross Burton <ross.burton@arm.com>
|
| |
|
|
| |
Signed-off-by: Khem Raj <khem.raj@oss.qualcomm.com>
|
| |
|
|
| |
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
|
|
| |
This is different to the setting in OE-Core and is causing failures with
yocto-check-layer. The referenced bug was closed a while ago so the issue
is likely fixed, remove the conflicting definition.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
| |
|
|
|
|
|
|
| |
This is different to the setting in OE-Core and is causing failures with
yocto-check-layer. The referenced bug was closed a while ago so the issue
is likely fixed, remove the conflicting definition.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
| |
|
|
| |
Signed-off-by: Martin Jansa <martin.jansa@gmail.com>
|
| |
|
|
|
|
|
| |
compiler-rt recipe is now in oe-core and the described issue doesn't
happen anymore.
Signed-off-by: Martin Jansa <martin.jansa@gmail.com>
|
| |
|
|
|
|
|
| |
It probably wasn't actively used for a while, it was last user of LLVMVERSION
variable which wasn't set correctly and was removed in previous commit
Signed-off-by: Martin Jansa <martin.jansa@gmail.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The recipes were removed from this layer and this version doesn't match
with the clang version in oe-core (currently 21.1.7).
Better to drop this variable now as it's just confusing and if someone
needs to find a path to clang lib files like:
recipe-sysroot/usr/lib/clang/18.1.6/lib/linux
we need to use some other variable to do that.
scripts/devtool-clang.sh still references this variable, but I guess
the script isn't used anymore.
Signed-off-by: Martin Jansa <martin.jansa@gmail.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
using LTO.
Referring to lto.inc in openembedded-core, when using lto, the LTO variable is added to SELECTED_OPTIMIZATION.
In this case, with the clang toolchain, -fuse-ld=lld is added when using -flto,
but the -fuse-ld=lld option is sufficient as a linker flag.
If -fuse-ld=lld is present in CFLAGS, an unused-command-line-argument warning occurs.
In recipes that treat this warning as an error, the build fails.
aarch64-yoe-linux-clang: error: argument unused during compilation: '-fuse-ld=lld' [-Werror,-Wunused-command-line-argument]
See more details https://errors.yoctoproject.org/Errors/Details/886599
Signed-off-by: mark.yang <mark.yang@lge.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
bluez5 is forced to use the bfd linker.
https://github.com/kraj/meta-clang/commit/ee218b7bc9c4b43d66e56c9cde5405318d8d5deb
systemd-boot is forced to use the bfd linker.
https://github.com/openembedded/openembedded-core/commit/a157b2f9d93428ca21265cc860a3b58b3698b3aa#diff-02f955f0bb176d6a95cf74d4fa6b499d0318f7877cbe3c525936b09a4dc8f3ce
For LTO, the lld linker is required, but if using the bfd linker, the LLVMgold.so plugin must be used. However, the gold linker is no longer used.
```
usr/bin/aarch64-yoe-linux/aarch64-yoe-linux-ld.bfd: recipe-sysroot-native/usr/bin/aarch64-yoe-linux/../lib/LLVMgold.so: error loading plugin: /home/markyang/workspace/yoe-distro/build/tmp/work/cortexa72-yoe-linux/systemd-boot/257.8/recipe-sysroot-native/usr/bin/aarch64-yoe-linux/../lib/LLVMgold.so: cannot open shared object file: No such file or directory
````
openembedded-core(be5856616) forces the use of the bfd linker for zlib.
zlib does not build with lld, keep it until https://github.com/madler/zlib/pull/936
is addressed
When using LTO with clang, it is recommended to use the lld linker.
If it use the bfd linker, it need to use the LLVMgold.so plugin,
but the gold linker has been deprecated.
See more details https://errors.yoctoproject.org/Errors/Details/886598
```
error loading plugin: TOPDIR/build/tmp/work/cortexa72-yoe-linux/zlib/1.3.1/recipe-sysroot-native/usr/bin/aarch64-yoe-linux/../lib/LLVMgold.so: cannot open shared object file: No such file or directory
```
Signed-off-by: mark.yang <mark.yang@lge.com>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When enabling LTO in the clang toolchain, .o files are generated in LLVM IR bitcode format instead of ELF format.
In this case, dwarfsrcfiles fails in do_package.
See more details at: https://errors.yoctoproject.org/Errors/Details/886597
```
dwarfsrcfiles: TOPDIR/build/tmp/work/cortexa72-yoe-linux/libcxx/21.1.4/package/usr/lib/libc++experimental.a: not a valid ELF file
```
As already discussed on the mailing list https://www.mail-archive.com/yocto@lists.yoctoproject.org/msg15501.html,
setting `INHIBIT_PACKAGE_DEBUG_SPLIT = "1"` disables debug splitting.
It is not desirable to disable debug splitting just to avoid the dwarfsrcfiles error.
A simple solution, as Khem Raj suggested in the lto.inc file of openembedded-core,
is to add the `-ffat-lto-objects` flag to the LTO settings.
This keeps both IR bitcode and ELF format, preventing the error.
Signed-off-by: mark.yang <mark.yang@lge.com>
|
| |
|
|
| |
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
Its in main recipe now
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
environment has the paths to it.
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
|
|
| |
This has adverse effect now with clang-21 on riscv64 qemu
where opensbi ends up not being found during boot and qemu
getting stuck in early boot.
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
|
|
| |
Clang-21 and OE-core move of clang has made several packages work
either out of box or the needed patches have moved to recipes proper
therefore remove them here.
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
This variable is not used anymore after clang migration to OE-core
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
This variable is dead-code after clang moved to OE-core
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
It was not displaying the status correctly and wad disabled
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
| |
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
It can now build with default lld options
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
| |
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
| |
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
| |
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
clang runtime recipe set them rightly in core
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
| |
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
|
| |
This is better fixed in recipes if needed now that
clang is a core recipe.
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
This is gone from OE-core so good riddance
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
Its working again
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
|
| |
These are not supported by clang, but now they are handled in core
tune files.
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
| |
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
Set TOOLCHAIN_NATIVE = "clang" in disrto conf files not here
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
|
| |
This patch is not needed anymore with clang 20+ and samba 4.19+
it build fine with both linkers ( LLD and Gnu LD)
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
This should be part of default tune infra in core
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
| |
Its moved into klibc.bbclass in meta-initramfs
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
|
|
| |
Some of these settings are moved to OE-core and some are deferred to
distros where it belongs. This makes meta-clang compatible with YP
again.
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|
| |
|
|
|
|
|
| |
They should be set by distros not by meta-clang
since distros decide to use clang as system compiler
Signed-off-by: Khem Raj <raj.khem@gmail.com>
|