Richard Purdie [Wed, 15 Oct 2025 22:24:42 +0000 (23:24 +0100)]
perf: Tweak reproducibility fix
Instead of building libperf.a, we should run install_headers as with the other
libraries. Hopefully this resolves the remaining race issue around headers.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Khem Raj [Tue, 14 Oct 2025 05:52:29 +0000 (22:52 -0700)]
vte: Depend on system provided fmt package
vte 0.82+ has started to package fmt as a subproject if this is not found
on system, the bundled version however, does not work with clang on
32bit machines, since it is 11.0 and there are fixes in newer version
needed to work with clang
Fixes
../sources/vte-0.82.1/subprojects/fmt/include/fmt/format.h:752:35: error: call to function 'free' that is neither visible in the template
Richard Purdie [Wed, 15 Oct 2025 16:50:52 +0000 (17:50 +0100)]
patchelf: Update 0.18.0 -> 0.18.0+git
Patchelf hasn't released since 2023 but does have fixes on its master branch. We've been
seeing segfaults on relocated qemu-img binaries from qemu-system-native in some cases
and using an updated patchelf does seem to avoid these.
Richard Purdie [Wed, 15 Oct 2025 14:49:24 +0000 (15:49 +0100)]
distro/defaultsetup: Add buildstats by default
This has been the default in poky's local.conf.sample since forever. It was missing
during the migrtion to bitbake-setup and that created a few failures. We've fixes most
of the places but found a new one in the performance tests.
Having these available is useful for debugging and doesn't really add much overhead
to the build.
We could push this over into the poky DISTRO however I've been wanting to
try and reconcile things where possible so putting this into defaultsetup feels
like the right choice to me. Distros can still override as they would the other
classes in INHERIT_DISTRO.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
rust-target-config: PPC64 targets require explicit ABI selection to avoid build failures with rustc.
Without a specified ABI, rustc panics with the following error:
| thread 'rustc' panicked at compiler/rustc_codegen_ssa/src/back/metadata.rs:394:21:
| No ABI specified for this PPC64 ELF target.
As noted in the upstream changes:
If the flags do not correctly indicate the ABI,
linkers such as ld.lld assume that the ppc64 object files are always ELFv2,
which leads to broken binaries if ELFv1 is used for the object files.
Because of this, it is now required to explicitly specify the ABI for PPC64 targets
using one of the following:
"elfv1" => EF_PPC64_ABI_ELF_V1,
"elfv2" => EF_PPC64_ABI_ELF_V2,
If no ABI is specified, the Rust compiler will panic with the error:
No ABI specified for this PPC64 ELF target
To address this:
- Set 'elfv2' for powerpc64le (little-endian), which mandates ELFv2 ABI.
- Set 'elfv1' for powerpc64 (big-endian), which defaults to ELFv1 ABI.
Signed-off-by: Deepesh Varatharajan <Deepesh.Varatharajan@windriver.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Khem Raj [Tue, 14 Oct 2025 00:08:33 +0000 (17:08 -0700)]
libpam: Fix build with LLD linker
LLD is strict about versioned symbols unlike BFD linker, it flags
undefined ones, Allow undefined symbols like BFD linker
Fixes
| riscv64-yoe-linux-ld.lld: error: version script assignment of 'global' to symbol 'pam_sm_acct_mgmt' failed: symbol not defined
| riscv64-yoe-linux-ld.lld: error: version script assignment of 'global' to symbol 'pam_sm_chauthtok' failed: symbol not defined
| riscv64-yoe-linux-ld.lld: error: version script assignment of 'global' to symbol 'pam_sm_close_session' failed: symbol not defined
| riscv64-yoe-linux-ld.lld: error: version script assignment of 'global' to symbol 'pam_sm_open_session' failed: symbol not defined
| riscv64-yoe-linux-clang: error: linker command failed with exit code 1 (use -v to see invocation)
Signed-off-by: Khem Raj <raj.khem@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Dmitry Baryshkov [Sun, 12 Oct 2025 15:08:04 +0000 (18:08 +0300)]
x264: switch to PACKAGECONFIG
Switch to PACKAGECONFIG in order to make x264 options configurable.
Enable FFmpeg (by default) and OpenCL (if enabled by the distro). Pick
up the patch to fix building with the latest FFmpeg.
Khem Raj [Sat, 11 Oct 2025 23:05:35 +0000 (16:05 -0700)]
llvm/clang: Upgrade to 21.1.3 release
brings following fixes
* 450f52eec88f Bump version to 21.1.3.
* 05b5090e961f Port 5b4819e to release/21.x
* 2cb08dbb39b1 [compiler-rt][sanitizer] fix msghdr for musl (#136195)
* 6e687cbe0dd3 [SPARC] Prevent meta instructions from being inserted into delay slots (#161111)
* f8151a1d2c86 [clang] [Headers] Don't use unreserved names in avx10_2bf16intrin.h (#161824)
* 9ee4ac8a8359 [clang][SPARC] Pass 16-aligned structs with the correct alignment in CC (#155829)
* 220bac16a417 [Hexagon] Add opcode V6_vS32Ub_npred_ai for offset validity check (#161618)
* a867bd53e861 [clang][PAC] Don't try to diagnose use of pointer auth on dependent types #159505 (#159859)
* 41e817a1d1f4 release/21.x: [clang-format] Fix bugs in annotating arrows and square brackets (#160973)
* 559d966bcb54 [Mips] Fix atomic min/max generate mips4 instructions when compiling for mips2 (#159717)
* d1e2f8916128 [LLD] [COFF] Fix symbol names for import thunks (#160694)
* 0060034c6a0b [analyzer] Revert #115918, so empty base class optimization works again (#157480)
* 31e4363ba9c2 [NVPTX] Disable relative lookup tables (#159748)
* 74cb34a6f51a Bump version to 21.1.3
Currently, the definitions of MIRRORS and PREMIRRORS made by
test_yocto_source_mirror() are overwritten when run on the autobuilder:
* MIRRORS = "" is extended by mirrors.bbclass to its usual value.
* PREMIRRORS = "* dl.yp.org" is overwritten by the autobuilder.conf
config fragment to ""
Despite this unexpected configuration, the test passes because after
failing to find a PREMIRROR (empty), failing to download the normal
SRC_URI (because of BB_ALLOWED_NETWORKS), the fetcher tries the MIRRORS
which have by default download.yoctoproject.org. For example, on a
failed meta-oe-mirror test[0], we can see a lot of warnings:
"stdio: WARNING: ... do_fetch: Failed to fetch URL https://... attempting MIRRORS if available"
By using the ":forcevariable" override, test_yocto_source_mirror() makes
sure the correct value is set for MIRRORS and PREMIRRORS (whatever is
the configuration).
Dmitry Baryshkov [Fri, 10 Oct 2025 11:59:51 +0000 (14:59 +0300)]
mesa: sort out driver lists
Sort the lists of VUKAN_DRIVERS, GALLIUMDRIVERS and driver-related
PACAKGECONFIG entries, making it easier to add new drivers or to find
out the correct dependencies for the driver.
Dmitry Baryshkov [Fri, 10 Oct 2025 11:59:50 +0000 (14:59 +0300)]
mesa: be more explicit about gallium-llvm dependency
Historically mesa recipe had separate variables for drivers that
depended on gallium-llvm PACKAGECONFIG. Since that time we started
listing other dependencies explicitly (e.g. libclc). Drop the
intermediate variables and make the dependency on gallium-llvm more
explicit.
Dmitry Baryshkov [Fri, 10 Oct 2025 11:59:48 +0000 (14:59 +0300)]
mesa: add opencl -> clang build dependency
With the commits 448f4a84cb22 ("llvm: add recipe for just the LLVM
libraries") and d76dc362c8e1 ("clang: use llvm recipe") the 'llvm'
dependency inside mesa.inc does no longer pull in the clang libraries,
failing RustiCL build as it can not find Clang libaries.
Add direct dependency on the clang in order to fulfill build-time deps.
Fixes: d76dc362c8e1 ("clang: use llvm recipe") Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Peter Marko [Fri, 10 Oct 2025 20:52:00 +0000 (22:52 +0200)]
binutils: patch CVE-2025-11083
Pick patch per link in NVD report.
Signed-off-by: Peter Marko <peter.marko@siemens.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Peter Marko [Fri, 10 Oct 2025 20:51:59 +0000 (22:51 +0200)]
binutils: patch CVE-2025-11082
Pick patch per link in NVD report.
Signed-off-by: Peter Marko <peter.marko@siemens.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
When building cargo-c with DEBUG mode, QA checks fail due to
absolute paths from the build environment (TMPDIR) being embedded into
the debug symbols/binaries. This causes issues such as:
ERROR: do_package_qa: File /usr/bin/.debug/cargo-capi contains reference to TMPDIR [buildpaths]
ERROR: do_package_qa: File /usr/bin/.debug/cargo-cinstall contains reference to TMPDIR [buildpaths]
ERROR: do_package_qa: File /usr/bin/.debug/cargo-cbuild contains reference to TMPDIR [buildpaths]
ERROR: do_package_qa: File /usr/bin/.debug/cargo-ctest contains reference to TMPDIR [buildpaths]
To fix this, pass an additional -ffile-prefix-map option to DEBUG_PREFIX_MAP to
ensure that paths under `${CARGO_HOME}` are remapped to `${TARGET_DBGSRC_DIR}`.
This ensures debug info is reproducible and does not leak host-specific paths.
Ryan Eatmon [Tue, 7 Oct 2025 22:43:49 +0000 (17:43 -0500)]
u-boot: Make sure the build dir is unique for each UBOOT_CONFIG
Each UBOOT_CONFIG entry is run in a different directory under ${B} so
that the files can be generated, compiled, and installed differently
from each other. Currently that unique directory name was just the
defconfig used for each UBOOT_CONFIG.
One potential conflict arises when you want build the same defconfig
twice, but pass in different make options. Then we get directory
collision. Simple fix is to include both the defconfig name and the
UBOOT_CONFIG type in the directory name.
This change has the potential to be backwards breaking if a layer is
using the UBOOT_CONFIG flow and overriding/appending any of the do_*
shell functions. Each of those will either need to change to using:
${B}/${config} -> ${B}/${config}-${type}
or for append functions they can use the new variable in the parent
function:
${B}/${config} -> ${B}/${builddir}
Signed-off-by: Ryan Eatmon <reatmon@ti.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Randy MacLeod [Tue, 7 Oct 2025 20:46:13 +0000 (16:46 -0400)]
gawk: disable persistent memory allocator due to licensing
In gawk-5.2, a feature that allows gawk to preserve memory between runs:
https://www.gnu.org/software/gawk/manual/html_node/Persistent-Memory.html
was added. The files that implement this, support/pma.[ch], are licensed
under the AGPL3 which some entities perfer to avoid. Force people to knowingly
opt into using this feature and license using:
PACKAGECONFIG:append:pn-gawk = " pma-if-64bit"
where pma is an abbreviation for persistent memory allocator and the
"-if-64bit" suffix is a indicator to users that the feature only works
for 64 bit targets. Also add AGPL to LICENSE and LICENSE:${PN}, when using pma.
Correct the license to be AGPL-3.0-or-later. There hasn't been a change in
the license terms, at least for main.c, haven't changed significantly in
the last 15 years:
License-Update: Reflects conditional AGPL use and more as described above.
Testing requires a non-root account and following the example in the link above:
$ truncate -s <size> data.pma
$ chmod 0600 data.pma
$ GAWK_PERSIST_FILE=data.pma gawk 'BEGIN { print ++i }'
1
$ GAWK_PERSIST_FILE=data.pma gawk 'BEGIN { print ++i }'
2
$ GAWK_PERSIST_FILE=data.pma gawk 'BEGIN { print ++i }'
3
This works on qemu[x86-|arm|riscv]64 but not on qemu[arm|x86] where the
--enable-pma is ignored because there is a requirement and build-time test
for 8 byte void pointers in m4/pma.m4:
if test "$SKIP_PERSIST_MALLOC" = no && test $ac_cv_sizeof_void_p -eq 8
Finally, remove an old comment about GPLv2, GPLv3 versions of gawk
since this is no longer important as the GPLv2 version is not maintained.
Signed-off-by: Randy MacLeod <Randy.MacLeod@windriver.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Peter Marko [Tue, 7 Oct 2025 20:13:15 +0000 (22:13 +0200)]
go: upgrade 1.25.1 -> 1.25.2
Upgrade to latest 1.25.x release [1]:
$ git --no-pager log --oneline go1.25.1..go1.25.2 bed6c81c2d (tag: go1.25.2) [release-branch.go1.25] go1.25.2 2612dcfd3c [release-branch.go1.25] archive/tar: set a limit on the size of GNU sparse file 1.0 regions 90f72bd500 [release-branch.go1.25] encoding/pem: make Decode complexity linear e0f655bf3f [release-branch.go1.25] encoding/asn1: prevent memory exhaustion when parsing using internal/saferio 100c5a6680 [release-branch.go1.25] net/http: add httpcookiemaxnum GODEBUG option to limit number of cookies parsed f0c69db15a [release-branch.go1.25] crypto/x509: improve domain name verification 9fd3ac8a10 [release-branch.go1.25] net/url: enforce stricter parsing of bracketed IPv6 hostnames 5d7a787aa2 [release-branch.go1.25] net/textproto: avoid quadratic complexity in Reader.ReadResponse 930ce220d0 [release-branch.go1.25] crypto/x509: mitigate DoS vector when intermediate certificate contains DSA public key 6a057327cf [release-branch.go1.25] net/mail: avoid quadratic behavior in mail address parsing 66f6feaa53 [release-branch.go1.25] spec: revert "update spec date to match release date" d6f2741248 [release-branch.go1.25] spec: update spec date to match release date 28ac8d2104 [release-branch.go1.25] net/http: avoid connCount underflow race 06993c7721 [release-branch.go1.25] context: don't return a non-nil from Err before Done is closed 0b53e410f8 [release-branch.go1.25] debug/pe: permit symbols with no name 7735dc90ed [release-branch.go1.25] cmd/compile: don't rely on loop info when there are irreducible loops 205d086595 [release-branch.go1.25] crypto/tls: quote protocols in ALPN error message 16fdaac4b1 [release-branch.go1.25] sync/atomic: correct Uintptr.Or return doc f3dc4aac0b [release-branch.go1.25] runtime: initialise debug settings much earlier in startup process 79c3081b4b [release-branch.go1.25] internal/poll: don't call Seek for overlapped Windows handles b816c79658 [release-branch.go1.25] lib/fips140: re-seal v1.0.0 90de3b3399 [release-branch.go1.25] crypto/internal/fips140: remove key import PCTs, make keygen PCTs fatal bec452a3a2 [release-branch.go1.25] crypto/internal/fips140: update frozen module version to "v1.0.0" 57bd28ab7f [release-branch.go1.25] crypto/internal/fips140/ecdsa: make TestingOnlyNewDRBG generic f75bcffa4a [release-branch.go1.25] os: set full name for Roots created with Root.OpenRoot 7d570090a9 [release-branch.go1.25] os: fix Root.MkdirAll to handle race of directory creation be61132165 [release-branch.go1.25] cmd/compile: export to DWARF types only referenced through interfaces a86792b169 [release-branch.go1.25] net: skip TestIPv4WriteMsgUDPAddrPort on plan9 879e3cb5f7 [release-branch.go1.25] runtime: lock mheap_.speciallock when allocating synctest specials
Signed-off-by: Peter Marko <peter.marko@siemens.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Peter Marko <peter.marko@siemens.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Mon, 13 Oct 2025 11:32:20 +0000 (12:32 +0100)]
perf: Improve build race/reproducibulity fixes
The include options shouldn't be needed with the make fix but issues still remained.
Looking at the logs, it looks like these are from other header directories and
we need to run "make install_headers" for all of the sub components before starting
the main build.
Update the workaround to do that for each component with internal header copies.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ryan Eatmon [Fri, 10 Oct 2025 16:08:42 +0000 (11:08 -0500)]
linux-firmware: Set FILES to ""
Now that all of the current firmware has either been moved into
sub-packages or removed due to licensing the base package is now
empty. Going forward we would like to keep the base package empty
and force the version updates to bin any new firmwares into
sub-packages.
Signed-off-by: Ryan Eatmon <reatmon@ti.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ryan Eatmon [Fri, 10 Oct 2025 16:08:41 +0000 (11:08 -0500)]
linux-firmware: Remove unlicensed firmware
Create a new REMOVE_UNLICENSED variable and do_install() logic to
allow for marking certain firmware files as something to remove and not
package up.
This comes from the fact that the WHENCE file has a number of firmware
entries that have no license information or very questionable
declarations as to what the license is for the firmware.
Signed-off-by: Ryan Eatmon <reatmon@ti.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ryan Eatmon [Fri, 10 Oct 2025 16:08:40 +0000 (11:08 -0500)]
linux-firmware: Move all firmware into sub-packages
A lot of the firmware has already been broken out into sub-packages, but
a good number were still lingering in the base linux-firmware package.
Move all of the remaining firmware into sub-packages based on the
driver name in the WHENCE file or into existing sub-packages where a
file or two were missed.
Signed-off-by: Ryan Eatmon <reatmon@ti.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add a zsh-completion-pkgs image feature to install *-zsh-completion
packages into an image. This is similar to the existing
bash-completion-pkgs feature.
Suggested-by: Peter Kjellerstedt <peter.kjellerstedt@axis.com> Signed-off-by: Adam Nilsson <Adam.X.Nilsson@axis.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Steve Sakoman [Thu, 9 Oct 2025 02:31:23 +0000 (19:31 -0700)]
oeqa/sdk/cases/autotools.py: use gnu mirror instead of main server
ftp.gnu.org is the main server of the GNU project, however download speed
can vary greatly based on one's location.
Using ftpmirror.gnu.org should redirect the request to the closest up-to-date mirror,
which should result sometimes in significantly faster download speed, depending
on one's location. This should also distribute the traffic more across the mirrors.
This information was sourced from https://www.gnu.org/prep/ftp.html
Signed-off-by: Steve Sakoman <steve@sakoman.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Steve Sakoman [Thu, 9 Oct 2025 02:31:22 +0000 (19:31 -0700)]
oeqa/selftest/cases/meta_ide.py: use gnu mirror instead of main server
ftp.gnu.org is the main server of the GNU project, however download speed
can vary greatly based on one's location.
Using ftpmirror.gnu.org should redirect the request to the closest up-to-date mirror,
which should result sometimes in significantly faster download speed, depending
on one's location. This should also distribute the traffic more across the mirrors.
This information was sourced from https://www.gnu.org/prep/ftp.html
Signed-off-by: Steve Sakoman <steve@sakoman.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ryan Eatmon [Tue, 7 Oct 2025 16:22:56 +0000 (11:22 -0500)]
u-boot: Add specifying make options as part the config looping
There is a need to generate alternative versions of the uboot files
using the existing config looping system, but we need to add additional
settings to the make call rather simply specifying a different config.
Specifically we have two use cases:
1) We want to sign the same uboot files with two different keys where
the key will be passed on the make call.
2) We want to include the alternative defconfigs from a different
repository and need to add the path to this new location on the make
command line.
This introduces a fourth value for the UBOOT_CONFIG settings:
config,images,binary,make_opts
The values are placed into a new generated variable
UBOOT_CONFIG_MAKE_OPTS which is a '?' separated list since space can be
present if you need to specify multiple options. This is handled by
changing IFS in the shell code when looping over the variable.
Additionally, add in a new variable UBOOT_MAKE_OPTS which is added to
the make calls in the various do_compile functions that do the actual
compiling.
Signed-off-by: Ryan Eatmon <reatmon@ti.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Postactions are not part of the tests but allow to retrieve useful data
from the target. They try to do this using SSH, but this can fail when
no SSH server is present on the target. Ignore these fails.
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
oeqa: runtime: Ignore SSH errors during setup and tear down
Tests using SSH will fail when no SSH server is present on the target.
These tests are disabled in these cases, by being marked with a
dependency on ssh.SSHTest.test_ssh, which in turns has a dependency on
having either dropbear or openssh-sshd in the image.
But setUpClass() and tearDownClass() functions are always executed, even
on tests failing the dependency checks, leading to unexpected failed
tests.
Ignoring SSH errors in setup and tear down allows to avoid these test
errors.
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
oeqa: target: ssh: Fail on SSH error even when errors are ignored
Most tests running SSH commands ask for no error to be raised when the
returned status is not 0. As run() will return this status, they may
later use its value to do a similar check on their own, or completely
ignore it. But most of the tests do not check if the non-zero status is
caused by a fail of the command run on the target or by a fail of SSH
itself.
This can lead to confusion when the error does not come from the command
executed on the target but from SSH itself: test might wrongfully be
marked as PASSED or might fail with incoherent errors.
As SSH errors are always reported with exit code 255, we can easily
filter these.
Modify OESSHTarget.run() behaviour so an AssertionError is raised on SSH
failures, even when ignore_status parameter is True. Still allow to
explicitly ignore this error for the rare cases where this can be
needed.
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Yi Zhao [Tue, 7 Oct 2025 14:51:32 +0000 (22:51 +0800)]
tcl8: upgrade 8.6.16 -> 8.6.17
ChangeLog:
2024-12-16 (bug) [63449c] [namespace children] doesn't match non-glob
patterns below the global namespace
2025-01-06 (bug) [fc3509] Better error-message than "interpreter uses
an incompatible stubs mechanism"
2025-01-19 tzdata updated to Olson's tzdata2025a
2025-01-28 (bug) [4f0b57] Win: [exec] now works on App Execution
Aliases.
2025-01-28 (bug) [4e2c8b] Win: [auto_execok] handles larger set of
shell commands.
2025-03-06 (bug) [ba68d1] errorline from [interp eval], interp-26.9
2025-03-23 tzdata updated to Olson's tzdata2025b
2025-04-11 (bug) [fd8341] Tcl_InitStubs compatibility for 9.1, better
error-handling
2025-05-05 (bug) [42d14c] Fix scan with long mantissa. Ex.: scan
"1.[string repeat 1 191]e-321" %g
2025-06-18 (bug) [4f338b] add missing Tcl_CloseEx docs
2025-06-24 (bug) [ecf35c] Correct nested handling of return option
-options
2025-06-25 (bug) [ecafd8] Euro/Tail-sign missing from cp864 encoding
2025-07-03 (bug) [6b0f77] gcc 14 breaks configure test for bigendian
leading to broken floating point
2025-07-16 (bug) [c9f052] prevent overflow crash in Tcl_SplitList().
2025-07-21 (bug) [61c01e] Flawed ref counts in filesystem
implementation for Windows led to use-after-free
2025-08-12 (new) dde => 1.4.5
2025-08-12 (bug) [992f94] avoid misaligned pointers in macOS file
attribute functions
Set LC_ALL and LANG to en_US.UTF-8 when running ptest since the test
cases now include more encodings than just ASCII[1].
Also, add rdepends on locale-base-en-us and tzdata for ptest package,
as they are required for running ptest.
Khem Raj [Mon, 6 Oct 2025 21:48:48 +0000 (14:48 -0700)]
ghostscript: Do not treat declaration-after-statement warning as error
This option is added by ghostscript and is passed down to its modules
e.g. brotli, brotli does expect c99 or newer standard and hence uses
declarations after statement. This option causes compiler e.g. clang
to find this warning and treat it as error on 32bit builds
./brotli/c/dec/decode.c:440:12: error: mixing declarations and code is incompatible with standards before C99 [-Werror,-Wdeclaration-after-statement]
440 | uint32_t __fastload_table = (*table);
| ^
1 error generated.
recipeutils/get_recipe_upstream_version: pass ud.name instead of 'default'
While all but the osc fetcher ignore the third parameter of their
latest_revision implementation, 'default' isn't a valid name in general.
Since commit 2515fbd10824 ("fetch: Drop multiple branch/revision support
for single git urls") in bitbake a fetcher only handles a single
branch/revision and the only sensible thing to pass is `ud.name`.
OpenSSL 3.5.4 is a security patch release. The most severe CVE fixed in this release is Moderate.
This release incorporates the following bug fixes and mitigations:
* Fix Out-of-bounds read & write in RFC 3211 KEK Unwrap. (CVE-2025-9230)
* Fix Timing side-channel in SM2 algorithm on 64 bit ARM. (CVE-2025-9231)
* Fix Out-of-bounds read in HTTP client no_proxy handling. (CVE-2025-9232)
* Reverted the synthesised OPENSSL_VERSION_NUMBER change for the release builds, as it broke some exiting applications that relied on the previous 3.x semantics, as documented in OpenSSL_version(3).
OpenSSL 3.5.3 is a bug fix release.
This release incorporates the following bug fixes and mitigations:
* Added FIPS 140-3 PCT on DH key generation.
* Fixed the synthesised OPENSSL_VERSION_NUMBER.
* Removed PCT on key import in the FIPS provider as it is not required by the standard.
Signed-off-by: Peter Marko <peter.marko@siemens.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
selftest/meta_ide: source the environment first, then change to the sources directory
This too used to work by coincidence: sourcing the environment
quietly failed without changing to the build directory, but
now that it works properly, things should be done in correct order.
Signed-off-by: Alexander Kanavin <alex@linutronix.de> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Sourcing a script with arguments is a non-standard bash extension
and doesn't work with other shells (e.g. dash, which is used on
Debian and derivatives). This used to work by coincidence when
running against integrated poky repo without having to separately
specify where bitbake is, but no longer does.
Using set is a POSIX standard.
Signed-off-by: Alexander Kanavin <alex@linutronix.de> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Pass the BITBAKEDIR parameter to oe-init-build-env when called from the
environment-setup script. This fixes compatibility with bitbake-setup,
which places bitbake in a different directory structure than the
standard poky repository layout where bitbake is located alongside
oe-init-build-env.
The issue was discovered when running the oe-selftest test
DevtoolIdeSdkTests.test_devtool_ide_sdk_shared_sysroots, which failed
because the environment-setup script was not able to find bitbake.
File ".../openembedded-core/meta/lib/oeqa/selftest/cases/devtool.py",
line 2955, in test_devtool_ide_sdk_shared_sysroots
self.assertExists(cmake_native)
~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^
File ".../openembedded-core/meta/lib/oeqa/selftest/case.py", line 251,
in assertExists
raise self.failureException(msg)
AssertionError: 'Error: The bitbake directory (/tmp/devtoolqah9ndff2x/bitbake)
does not exist! Please ensure a copy of bitbake exists at this location or
specify an alternative path on the command line\n
.../build-st/tmp/sysroots/x86_64/usr/bin/cmake' does not exist
Signed-off-by: Adrian Freihofer <adrian.freihofer@siemens.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Grünert [Wed, 27 Aug 2025 06:49:40 +0000 (08:49 +0200)]
scripts/runqemu: raise an error when bitbake was not found
Running 'scrupts/runqemu' without bitbake in PATH causes the
following error:
```
Traceback (most recent call last):
File "/home/rg/temp_stuff/oe_2/./scripts/runqemu", line 1807, in main
config.check_args()
~~~~~~~~~~~~~~~~~^^
File "/home/rg/temp_stuff/oe_2/./scripts/runqemu", line 624, in check_args
s = re.search('^DEPLOY_DIR_IMAGE="(.*)"', self.bitbake_e, re.M)
File "/usr/lib/python3.13/re/__init__.py", line 177, in search
return _compile(pattern, flags).search(string)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^
TypeError: expected string or bytes-like object, got 'NoneType'
```
This patch adds a more helpful error message to inform the user that
bitbake was not found, e.g. because oe-init-build-env was not sourced.
This is an example of the new error message after the patch:
```
runqemu - ERROR - In order for this script to dynamically infer paths
kernels or filesystem images, you either need bitbake in your PATH
or to source oe-init-build-env before running this script.
Dynamic path inference can be avoided by passing a *.qemuboot.conf to
runqemu, i.e. `runqemu /path/to/my-image-name.qemuboot.conf`
Bitbake is needed to run 'bitbake -e', but it is not found in PATH. Please source the bitbake build environment.
```
CC: Richard Purdie <richard.purdie@linuxfoundation.org> CC: Alexander Kanavin <alex.kanavin@gmail.com> Signed-off-by: Richard Grünert <r.gruenert@pironex.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
runqemu: ensure that bitbake environment is either returned, or an exception is raised
This eliminates the other remaining code path where environment getter
returns 'nothing'. This and the previous patch were tested in a-full,
and no errors occurred [1], which means the code paths that make
use of the function returning nothing are never actually executed
and can be cleaned up (in the following patch).
The rationale is that if environment getter cannot obtain the environment,
it should report that and not sweep the issue under the carpet;
it's up to the caller to handle that situation, or make pre-emptive
checks that avoid calling the environment getter when it is bound to fail.
Ross Burton [Wed, 8 Oct 2025 14:26:20 +0000 (15:26 +0100)]
clang: consolidate LLVM_APPEND_VC_REV=OFF
Whilst the change to add TMPDIR to GIT_CEILING_DIRECTORIES should stop
LLVM from embedding git information into the recipes, also disable this
behaviour explicitly.
We do this because it's not just the sha of the source tree but also
the full URL of the repository, which would be an information leak if
an internal git mirror was being used.
Signed-off-by: Ross Burton <ross.burton@arm.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ross Burton [Tue, 7 Oct 2025 10:42:46 +0000 (11:42 +0100)]
clang: use llvm recipe
Change this recipe to build just clang and clang-tools-extra, using the
LLVM provided by the llvm recipe.
This adds an 'extra-tools' PACKAGECONFIG (enabled by default) that
controls whether to build the clang-tools-extra project. This includes
clang-tidy and clangd, but the compile time and size for these
components is not insignificant.
Add a patch from upstream to support using native prebuilt tools (such
as clang-tblgen) when building standalone.
Add a patch that is being worked on with upstream to not rebuild clang-
tblgen if it has already been provided. This saves a little build time,
but more importantly for us resolves a static linking/uninative problem.
Remove the dependency on clang-cross in non-native builds by just
depending on llvm-native and using the un-prefixed tools directly.
Signed-off-by: Ross Burton <ross.burton@arm.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ross Burton [Tue, 7 Oct 2025 10:42:45 +0000 (11:42 +0100)]
llvm: add recipe for just the LLVM libraries
Whilst it's convenient to build all of the LLVM project in one big
recipe, that's not ideal when we may just need LLVM on target and not
the rest.
Bring back a LLVM recipe that can be used by both clang (shortly) and
Rust (in the future)
Set the build type to MinSizeRel and DEBUG_LEVELFLAG to -g1 (instead of
the default, -g): the LLVM debug symbols are very large (several
gigabytes) and this reduces them to hundreds of megabytes.
Signed-off-by: Ross Burton <ross.burton@arm.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ross Burton [Tue, 7 Oct 2025 10:42:44 +0000 (11:42 +0100)]
clang: globally disable build-time RPATHs for reproducibility
Various bits of the LLVM project set the build RPATH in a way that means
we have non-deterministic binaries even though we should always be using
relative paths.
This clearly is not working as some of the binaries get rewritten on
install and have large string paddings that correlate with erased build
paths.
So that we don't have to disable RPATHs in every recipe, just do it once
in the common include file and remove the existing recipe-specific
assignments that are now not needed.
Signed-off-by: Ross Burton <ross.burton@arm.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ross Burton [Wed, 8 Oct 2025 14:02:24 +0000 (15:02 +0100)]
bitbake.conf: add TMPDIR to GIT_CEILING_DIRECTORIES
We export GIT_CEILING_DIRECTORIES=WORKDIR to ensure that git calls
inside the builds don't find oe-core when they're meant to be looking
for the git repository of the source code.
However, this breaks for recipes that use work-shared (such as llvm), as
their working directory is outside of WORKDIR.
Solve this by adding TMPDIR to the list as a final catch, but keeping
WORKDIR first so that git will stop sooner in the general case.
This solves reproduciblity problems in LLVM, where for example lld's
version string would contain the URL and commit hash of the poky repo
being built.
Signed-off-by: Ross Burton <ross.burton@arm.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Mon, 6 Oct 2025 13:21:12 +0000 (14:21 +0100)]
bitbake.conf,lib/configfragments: Use a new dedicated toolcfg.conf file
Rather than using auto.conf which already has established use in CI, or
local.conf which users expect to own/control, start writing "tooling"
controlled settings to a toolcfg.conf.
This frees CI to handle auto.conf as it wants, but avoids the tooling
breaking users local.conf files.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Mon, 6 Oct 2025 13:14:52 +0000 (14:14 +0100)]
oeqa/selftest/bblock/fitimage/sstatetests: Fix changing MACHINE during the test
With config fragments, changing MACHINE in the test like this no
longer works. Use the forcevarable override to allow it to work.
This also needs a tweak to bitbake to work correctly, sent
seperately. Whilst ugly, this avoids the need to start changing
config fragments within oeqa right now.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Mon, 6 Oct 2025 13:09:02 +0000 (14:09 +0100)]
oeqa/selftest: Fix single threaded race issue
oe-selftest sets up separate build directories to run the tests in.
To to this, environment paths pointing at the previous build directory
are updated. In the multi-threaded case this is fine as the thread is
destroyed and the parent remains unchanged but in the single threaded
case, the environment is broken afterwards. This can mean we try and access
a directory which is in the process of being deleted (e.g. by clobberdir).
Restore the environment afterwards regardless to ensure the single threaded
case doesn't try and access the build directory which is now being deleted.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
selftest/bblayers: maintain compatibility with integrated poky repo
This amends the recently merged commit that assumes the transition
to separate repositories has already happened and re-instatates
support for integrated poky.
Signed-off-by: Alexander Kanavin <alex@linutronix.de> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Yoann Congal [Fri, 3 Oct 2025 21:12:48 +0000 (23:12 +0200)]
oeqa/selftest/wic: fix PATH for wic.Wic2.test_extra_partition_plugin
Without importing PATH from the wic-tools recipes, the build host PATH
is used and this test may fail depending on tools (parted, dumpe2fs,
...) availability. This triggers build faillure on AB (e.g. [0])
To fix this, import PATH from wic-tools and ensure the original
environment is restored after.
Since this indent a block of code into a try/finally block, here is the
diff ignoring white spaces change:
diff --git a/meta/lib/oeqa/selftest/cases/wic.py b/meta/lib/oeqa/selftest/cases/wic.py
index bff3842305..bc99673d0d 100644
--- a/meta/lib/oeqa/selftest/cases/wic.py
+++ b/meta/lib/oeqa/selftest/cases/wic.py
@@ -1680,0 +1681,4 @@ INITRAMFS_IMAGE = "core-image-initramfs-boot"
+ oldpath = os.environ['PATH']
+ os.environ['PATH'] = get_bb_var("PATH", "wic-tools")
+
+ try:
@@ -1696,0 +1701,3 @@ INITRAMFS_IMAGE = "core-image-initramfs-boot"
+ finally:
+ os.environ['PATH'] = oldpath
+
Yoann Congal [Fri, 3 Oct 2025 21:12:47 +0000 (23:12 +0200)]
oeqa/selftest/wic: fix PATH for wic.Wic2.test_extra_partition_space
Without importing PATH from the wic-tools recipes, the build host PATH
is used and this test may fail depending on tools (parted, dumpe2fs,
...) availability. This triggers build faillure on AB (e.g. [0])
To fix this, import PATH from wic-tools and ensure the original
environment is restored after.
Since this indent a block of code into a try/finally block, here is the
diff ignoring white spaces change:
diff --git a/meta/lib/oeqa/selftest/cases/wic.py b/meta/lib/oeqa/selftest/cases/wic.py
index b1c318bd4e..34d844b90b 100644
--- a/meta/lib/oeqa/selftest/cases/wic.py
+++ b/meta/lib/oeqa/selftest/cases/wic.py
@@ -1331,0 +1332,4 @@
+ oldpath = os.environ['PATH']
+ os.environ['PATH'] = get_bb_var("PATH", "wic-tools")
+
+ try:
@@ -1366,0 +1371,2 @@
+ finally:
+ os.environ['PATH'] = oldpath
Peter Marko [Fri, 3 Oct 2025 18:25:33 +0000 (20:25 +0200)]
busybox: patch CVE-2025-46394
Pick commit mentioning this CVE.
Signed-off-by: Peter Marko <peter.marko@siemens.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Robert Tiemann [Thu, 2 Oct 2025 09:59:25 +0000 (11:59 +0200)]
udev-extraconf: Avoid slashes in mountpoint names
Devices with labels such as "Hello/World/Foo/Bar" cause mount.sh to
create the directory structure @MOUNT_BASE@/Hello/World/Foo/Bar. The
partition is mounted to the nested "Bar" directory. On device removal,
the directory structure is not cleaned up.
This commit replaces all forward slashes in partition labels by
underscores to avoid this edge case.
Signed-off-by: Robert Tiemann <rtie@gmx.de> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Robert Tiemann [Thu, 2 Oct 2025 09:59:24 +0000 (11:59 +0200)]
udev-extraconf: Speed up mount.sh
On devices with many partitions, running blkid without parameters can
be rather slow because all block devices are inspected:
$ time /sbin/blkid
real 0m0.474s
user 0m0.026s
sys 0m0.172s
versus
$ time /sbin/blkid /dev/mmcblk0p10
real 0m0.027s
user 0m0.002s
sys 0m0.018s
Plugging in a device with 5 partitions means that mount.sh is going to
be executed 5 times, and so will be blkid. In the real-world case
outlined above, this adds up to an overhead of about 2.3 seconds for
blkid alone.
This commit changes mount.sh so that the block device of interest is
passed directly to blkid such that blkid inspects only that device,
leading to significant speedup.
Signed-off-by: Robert Tiemann <rtie@gmx.de> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
The extra_partition plugin allows populating an extra partition with
files listed in the new IMAGE_EXTRA_PARTITION_FILES variable. The
implementation is similar to the bootimg_partition plugin.
This plugin provides an easy way to install files that are not part of
the rootfs, from the deploy directory.
mesa-demos: split info tools to a separate package
While the rest of programs inside mesa-demos are really "demos", several
*info utilities have separate value as they allow gathering information
about the running system in a manner similar to clinfo or vulkaninfo.
Split them into a separate package in order to allow picking them info
the images without picking up the rest of "demos".
Jon Mason [Wed, 1 Oct 2025 17:58:14 +0000 (13:58 -0400)]
tunes/arm: add support for newer arm64 cores
Add tunes for cores that have been present since GCC 14.1.0.
These have been present in meta-arm since mid-July, and have all been
verified with fvp-base and testimage. Of those present in meta-arm,
arch-armv8-9a.inc, arch-armv9-4a.inc, and arch-armv9-5a.inc have been
excluded from this commit, as there are currently issues compiling some
packages (but still present in meta-arm for those that want to use them
despite those issues).
Signed-off-by: Jon Mason <jon.mason@arm.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
b708aea0bc71 [SCEV] Don't perform implication checks with many predicates (#158652) 77a3b0eda361 [RISCV] Refactor RVV builtin code generation for reduce compilation time [NFC] (#154906) f14551dbc4e8 [RISCV] Reduce ManualCodeGen for RVV intrinsics with rounding mode. NFC dcc2c1c933be [RISCV] Reduce ManualCodeGen for segment load/store intrinsics. NFC e625a781211e [LLVM] Update CUDA ELF flags for their new ABI (#149534) 3e93017936b5 MC: Better handle backslash-escaped symbols (#158780) bc5e9a5e2009 [MC] Add parseSymbol() helper (NFC) (#158106) e2e5eb2f1cd9 [Loads] Check for overflow when adding MaxPtrDiff + Offset. 661c387fc2f1 release/21.x: [VPlan] Don't narrow op multiple times in narrowInterleaveGroups. f5c1b5206cbe [PowerPC] Avoid working on deleted node in ext bool trunc combine (#160050) db70369f400e [Clang] Fix an error-recovery crash after d1a80dea (#159976) 9c8736f348e0 [ELF] -r/--emit-relocs: Fix crash when processing .rela.text before .text (#156354) 15a31832eab7 [RISCV] Re-work how VWADD_W_VL and similar _W_VL nodes are handled in combineOp_VLToVWOp_VL. (#159205) f089fb21fffa [LoongArch] Fix MergeBaseOffset for constant pool index operand (#159336) 9eedaf5b1001 [VectorCombine] Fix scalarizeExtExtract for big-endian (#157962) 5af5cfb60d9e [clang-format] Handle C digit separators (#158418) 0174263ac214 Bump version to 21.1.2 ff039a98523f [RISCV] Support PreserveMost calling convention (#148214) 1a644bc50948 [X86] Fix assertion in AVX512 setcc combine due to invalid APInt mask width (#155775) 2c8cb316b5d2 [lldb][test] Only assert function name is in user-code on Darwin platforms f5ba88341e7c [lldb][test] TestTsanBasic.py: fix function name assertion faedeb1a2420 [lldb][Instrumentation] Set selected frame to outside sanitizer libraries (#133079) 54896838ca5e [lldb][Target] Clear selected frame index after a StopInfo::PerformAction (#133078) f4907049285c [Clang][Cygwin] Use correct mangling rule (#158404)
selftest/bblayers: do not assume integrated poky repo
The autobuilder is being transitioned to bitbake-setup which
sets up individual repositories that yocto is made of. Adjust
hardcoded revisions to match these repositories.
Signed-off-by: Alexander Kanavin <alex@linutronix.de> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
devtool: ide_sdk: pass BITBAKEDIR to oe-init-build-env
This fixes an issue where the generated install_and_deploy script is
unable to find the bitbake directory when run outside of the build
environment. This happens if the oe-selftest suite runs in a bitbake
environment that is bootstrapped by bitbake-setup.
AssertionError: Command '.../build-st/workspace/ide-sdk/cmake-example/
scripts/install_and_deploy_cmake-example-cortexa57' returned non-zero exit status 1:
Error: The bitbake directory (/tmp/devtoolqakq7kzgeo/bitbake) does not exist!
Please ensure a copy of bitbake exists at this location or specify an
alternative path on the command line
. /tmp/devtoolqakq7kzgeo/core-copy/oe-init-build-env
/home/adrian/bitbake-builds/poky-master-poky-with-sstate-distro_poky-altcfg-machine_qemuarm64/build-st
failed
Another reason this issue occurs with oe-selftests is that devtool
tests assume the full poky git repository is available. The setUpModule
function clones layer repositories, which for poky includes bitbake.
However, when using separate git repositories for bitbake and
openembedded-core, the bitbake directory is not preserved during layer
copying. While copying layers to allow modification during tests makes
sense, copying bitbake is less beneficial. Referring to the original
bitbake location is preferable, but cleaning up the devtool tests is
not part of this change.
Signed-off-by: Adrian Freihofer <adrian.freihofer@siemens.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
devtool: ide-sdk: use /bin/sh instead of /bin/bash
When generating the install and deploy script for IDEs, use /bin/sh
instead of /bin/bash. While this is not addressing a known issue,
using the more portable /bin/sh shell is preferable and avoids
requiring bash to be installed.
Signed-off-by: Adrian Freihofer <adrian.freihofer@siemens.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
lib/bbconfigbuild/configfragments: disable the previous builtin fragment when enabling a new one
There was a flaw in the logic that allowed multiple builtin fragments with
the same prefix to be enabled at the same time. The correct behaviour
is that only one of them should be enabled, and when enabling it
all previously enabled fragments should be removed.
The issues that this caused are further explained in
https://bugzilla.yoctoproject.org/show_bug.cgi?id=15987
[YOCTO #15987]
Signed-off-by: Alexander Kanavin <alex@linutronix.de> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Ross Burton [Mon, 29 Sep 2025 17:11:22 +0000 (18:11 +0100)]
linux-firmware: skip ldflags sanity check
Some of the firmware blobs, for example qcom/apq8016/modem.mbn, are
actually ELF files. We don't want to hold these to our exacting standard
for link flags because we're not building them and they don't run on the
target directly.
Note that as this check parses the output of ${OBJDUMP} -p it behaves
differently with binutils vs llvm. It looks like binutils bails early
as it doesn't know what the target architecture is, whereas LLVM is more
comprehensive in its support.
Signed-off-by: Ross Burton <ross.burton@arm.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
populate_sdk_ext: do not require CONF_VERSION (e.g. local.conf version) to be set
The version checks for local.conf/site.conf/bblayers.conf are all optional,
and aren't enforced (by insane class) if the versions aren't set.
As bitbake-setup writes out a blank local.conf, it doesn't put a version in it
either. Also, esdk bundle has a fixed set of layers and is not at risk of
needing to update its own local.conf.
The same condition is already in place for esdk's bblayers.conf for similar reasons.
Signed-off-by: Alexander Kanavin <alex@linutronix.de> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Changqing Li [Tue, 30 Sep 2025 08:00:55 +0000 (16:00 +0800)]
package.py: replace all files unconditionally when copy debug sources
This is for fixing reproducible issue for package like:
intel-speed-select-src. For intel-speed-select, one of the debug sources
is /usr/src/debug/intel-speed-select/1.0/include/linux/thermal.h,
file include/linux/thermal.h under ${S} (kernel-sources)
link file include/linux/thermal.h under ${B}, which link to ${S}/include/uapi/linux/thermal.h
During copy debug sources, sources under ${S} copied first, then sources
under ${B} is copied. mtime of ${S}/include/linux/thermal.h and
${S}/include/uapi/linux/thermal.h are decided by when it is fetched, so
it is not determinate, maybe same or different.
For cpio, if the in file is older than or the same as the exist file,
cpio will not replace the exist file with warning "cpio: xxx not created:
newer or same age version exists". And this will cause
intel-speed-select-src maybe not reproducible.
And option '-u' for cpio, first, this will make the copied file
determistic. Second, source files under ${B} should have higher priority
then ${S}, it may be generated during build, the target is more likely
to use this file.
Signed-off-by: Changqing Li <changqing.li@windriver.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Ryan Eatmon [Thu, 25 Sep 2025 19:21:15 +0000 (14:21 -0500)]
linux-firmware: Fix FILES for intel firmware
The 20250917 version is installing the firmware files into the
/usr/lib/firmware/intel directory and symbolic links to those files
under /usr/lib/firmware. FILES needs to be updated to correctly package
all of the files into the right packages. Without this we get a dependency
from the linux-firmware-iwlwifi-* and linux-firmware-qat packages back to
linux-firmware which means you have to install ALL of the drivers when you
depend on the specific package.
Signed-off-by: Ryan Eatmon <reatmon@ti.com> Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Ross Burton [Mon, 29 Sep 2025 17:11:25 +0000 (18:11 +0100)]
clang: build with MinSizeRel
This is the recommended build configuration upstream unless speed is the
ultimate goal. They say it's almost as fast as Release (-O3) but a lot
smaller on disk:
Ross Burton [Mon, 29 Sep 2025 17:11:24 +0000 (18:11 +0100)]
clang: fix reproducibility issue in LLVMConfig.cmake
The LLVMConfig.cmake file is mostly LLVM configuration, but it also
specifies if some Python modules (needed by the opt-viewer tool) were
present on the build host.
This is host contamination and a source of non-determinism, so remove it
from the installed file.
gn is a commonly used build tool to generate ninja files, used
in a lot of recipes e.g. chromium, qtwebengine, perfetto, hafnium etc.
these recipes come from different layers e.g. meta-qt6/meta-arm/meta-oe
since not all layers depend on meta-oe ( meta-arm does not ), its not
a possible option.
Given the usecases, putting it in core will benefit the ecosystem
and reduce some duplication. This recipe is taken from meta-arm
Disable maybe-initialized is a gcc specific option as error
Disable format warnings as errors, it adds -Wno-format explictly in
its build system
projects / thirdparty / openembedded / openembedded-core-contrib.git / log
