| Age | Commit message (Collapse) | Author |
|---|
|
This class provided default build flags but includes the base go class
which also provides default flags. Use a different mechanism to
extend the default flags without discarding `-trimpath` from the base
class.
Signed-off-by: Peter A. Bigot <pab@pabigot.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Commit c5dfc2586b41 (kernel.bbclass: run do_symlink_kernsrc before
do_patch) fixed a race between do_symlink_kernsrc and
do_populate_lic. However, I missed the fact that when
externalsrc.bbclass is in use, the do_patch task doesn't exist,
meaning that do_symlink_kernsrc now doesn't get run at all, breaking
the build.
We need this task to run regardless of whether do_patch exists or not,
so reinstate the configure->symlink_kernsrc dependency explicitly.
Fixes: c5dfc2586b41 (kernel.bbclass: run do_symlink_kernsrc before
do_patch)
Reported-by: Chanho Park <parkch98@gmail.com>
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
In contrast to 'touching' a file without parameters 'touch -a' can only be
performed by the file owner. In case of a shared sstate-cache served
locally (e.g. over NFS), where the user running bitbake is not the
owner of sstate-cache files, even if he has write access on group level,
the sstate_unpack_package task will fail.
Checking if the file is owned by the user before attempting to run
'touch -a' on it solves this.
Signed-off-by: Norman Stetter <norman.stetter@garz-fricke.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
It was pointed out that since the check for the existence of
KBUILD_DEFCONFIG knows where it is looking, that should be part of
the error/log message.
We only look in one location for the defconfigs, so add where we
checked to the message.
Signed-off-by: Bruce Ashfield <bruce.ashfield@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
RNG passthru has been enabled on all qemu machines but its being added
to each one of them, with this patch its turned into QB variables which
defaults to host passthru, yet it can be overridden if needed via
machine or config metadata if needed.
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
as otherwise when using it in combination with archiver.bbclass we would
have 2 methods of the same name but with different signatures
leading to various hard to understand exceptions
Signed-off-by: Konrad Weihmann <kweihmann@outlook.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Since the newly introduced image-artifact-names class is inherited at later
processing stage, individual IMAGE_NAME settings are overridden (hopefully
by accident instead on purpose).
Allow derived distributions define their own name schema by setting defaults
iwth question mark.
Signed-off-by: Jens Rehsack <sno@netbsd.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* to match other variables there like deployDir imageType
[YOCTO #12937]
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* they are used only once, we can use the value directly
* notice that .bin extension isn't part of the variable values
[YOCTO #12937]
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* just to make sure it looks like bash variable not bitbake variable in
run.do_* scripts
[YOCTO #12937]
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
* similar to kernel-artifact-names for other recipes/bbclasses which
need to use some deployed artifacts
* bitbake.conf: move IMAGE_BASENAME, IMAGE_VERSION_SUFFIX, IMAGE_NAME,
IMAGE_LINK_NAME variables
* image_types.bbclass: move IMAGE_NAME_SUFFIX variable
* currently IMAGE_NAME_SUFFIX is used only by image.bbclass,
image_types.bbclass and meta/recipes-core/images/build-appliance-image_15.0.0.bb
but if it's needed by some recipe which isn't itself an image, then
it's useful in bitbake.conf, e.g. we have a recipe for creating
VirtualBox appliances which combines .wic.vmdk with .ovf file to
create .zip with appliance, but for that we need the filename of
.wic.vmdk which now contains IMAGE_NAME_SUFFIX
https://github.com/webOS-ports/meta-webos-ports/blob/4980ce52a43ac6897657602810313af359f0b839/meta-luneos/recipes-core/images/luneos-emulator-appliance.inc#L24
* we were hardcoding .rootfs suffix where needed, but for quite long
time it's configurable with IMAGE_NAME_SUFFIX since:
commit 380ee36811939d947024bf78de907e3c071b834f
Author: Patrick Ohly <patrick.ohly@intel.com>
Date: Mon Mar 7 18:07:52 2016 +0100
image creation: allow overriding .rootfs suffix
and might not match with hardcoded .rootfs, so make it easier to
use IMAGE_NAME_SUFFIX where needed even without inheritting whole
image_types.bbclass
[YOCTO #12937]
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Add a package QA check to test if the recipe sets ALTERNATIVE variable
for any of its packages, and does not inherit update-alternatives class.
This causes the do_rootfs to fail since a proper alternative cannot be
created.
Signed-off-by: Andrey Zhizhikin <andrey.z@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
and bootimg-partition
Due to recent changes in bootimg-efi to include IMAGE_BOOT_FILES,
when both bootimg-partition and bootimg-efi occur in a single .wks
and IMAGE_BOOT_FILES are defined, files listed in IMAGE_BOOT_FILES
will be duplicated in both partition.
Since IMAGE_BOOT_FILES are crucial for bootimg-partition, but
optional for bootimg-efi, hence allowing bootimg-efi to have the option
to ignore it.
The new variable, IMAGE_EFI_BOOT_FILES, was added to help handle this
issue. Its basic usage is the same as IMAGE_BOOT_FILES.
Usage example:
${IMGDEPLOYDIR}/${IMAGE_BASENAME}-${MACHINE}.ext4;rootfs.img \
This commit is also squashed with the updated testcase to cover for
this change.
[YOCTO #14011]
Signed-off-by: Khairul Rohaizzat Jamaluddin <khairul.rohaizzat.jamaluddin@intel.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Some distros may not have local.conf, and use auto.conf or site.conf
alone to describe distro config metadata, therefore make code robust
to not fail eSDK builds for such setups
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Instead of inventing a new task to fetch the CVE data, use the existing
fetch task.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
If IMAGE_LINK_NAME is empty don't try to create the symlink.
[YOCTO #14042]
Signed-off-by: Diego Sueiro <diego.sueiro@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Signed-off-by: Khem Raj <raj.khem@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
For example, if you just run 'bitbake cve-update-db-native' in a clean
build system, |cve_tmp_file| won't exist yet.
Signed-off-by: Chris Laplante <chris.laplante@agilent.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Previously CVE_CHECK_DB_FILE / CVE_CHECK_DB_DIR was the same across
multiconfigs which led to a race condition wherein multiple
cve-update-db-native:do_populate_cve_db tasks could attempt to write to
the same sqlite database. This led to the following task failure:
Error executing a python function in exec_python_func() autogenerated:
The stack trace of python calls that resulted in this exception/failure was:
File: 'exec_python_func() autogenerated', lineno: 2, function: <module>
0001:
*** 0002:do_populate_cve_db(d)
0003:
File: '/mnt/data/agent/work/74f119cccb44f133/yocto/sources/poky/meta/recipes-core/meta/cve-update-db-native.bb', lineno: 103, function: do_populate_cve_db
0099: if year == date.today().year:
0100: cve_f.write('CVE database update : %s\n\n' % date.today())
0101:
0102: cve_f.close()
*** 0103: conn.commit()
0104: conn.close()
0105:}
0106:
0107:def initialize_db(c):
Exception: sqlite3.OperationalError: disk I/O error
Use a lockfile to ensure multiple tasks don't step over each other.
Signed-off-by: Chris Laplante <chris.laplante@agilent.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Commit 97b439469a45a089431ca9c31893288c855045f4 added a fallback
mechanism for getting the rootfs timestamp. However, it uses curly
braces around the variable name, which causes bitbake resolve the
variable reference, rather than the shell, so the git timestamp
never gets used. Fix the reference to restore the intent of
making it a fallback for when there is no git timestamp to
retrieve.
Signed-off-by: Matt Madison <matt@madison.systems>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The file path checks are passed an ELF object if the file is an ELF. It
doesn't make a lot of sense to load symlinks to ELFs as if they're in
the same package then the real file will be checked too.
This should speed up do_package_qa slightly as libraries won't be
scanned repeatedly.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
When the PR service is enabled a number of small changes may happen
to variables. In the do_package step a call to package_get_auto_pr
will end up setting PRAUTO and modifying PKGV (if AUTOINC is there).
PRAUTO is then used by EXTENDPRAUTO, which is then used to generate
PKGR.
Since this behavior typically happens BEFORE the BB_UNIHASH is
calculated for do_package, we need a way to defer the expansion
until after we have the unihash value.
Writing out the pkgdata files w/o AUTOPR and PKGV (AUTOINC) expanded
to placeholder values is the easiest way to deal with this. All other
variables are expanded as expected.
In the next task, typically do_packagedata, we will then use the
UNIHASH from the do_package to get the PR (AUTOPR) as well as
generate the AUTOINC replacement value (now PRSERV_PV_AUTOINC).
The do_packagedata then translates the placeholders to the final values
when copying the data from pkgdata to pkgdata-pdata-input.
Also update the prservice test case. With unihash, just changing the
do_package (via a _append) will not change the PR. So write the date
to a specific file that is incorporated into the unihash to ensure it
is always different for the test. Various assert messages were also
updated to make it easier to figure out where/why a problem occured.
Signed-off-by: Mark Hatle <mark.hatle@kernel.crashing.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
We just want to remove the first '-I' in $d, hence removing the
g' of sed avoid changing the real path which is include '-I'
We also add an anchor to the regex to only match -I at the start
of the path.
Signed-off-by: Yanfei Xu <yanfei.xu@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Instead of showing a filename under packages-split, show the package
name and filename inside the package.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Fix gnu-hash-style check that always returned True after commit [1]
leading to false positive presence of "GNU_HASH" in objdump output.
Fixes: 9ff90bf04a4c ("mips: Enable gnu-hash-style on glibc") [1]
Signed-off-by: Sumit Garg <sumit.garg@linaro.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
In the LICENSE_DIRECTORY each time an image recipe is built a new directory
is created to hold the image license manifests.
By creating a symlink to the most recent created image license manifest
directory, we make things easier for users to collect and pack the data.
Signed-off-by: Diego Sueiro <diego.sueiro@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
directories
This is based on the patch from Zhaolong Zhang <zhangzl2013@126.com>
[kernel-yocto: checksum indirect cfg and scc files]
While the recommended manner to share/reuse feature fragments is to
maintain them in a kernel-meta repository and track the changes via the
standard SRCREV fetcher mechanism, that method is not always practical
for small sets of features or for quick testing of changes.
These other flows use .scc files on the SRC_URI. It has been noted that
config fragments or other features indirectly included by those .scc
files will not trigger the kernel meta-data to be re-run and hence a
build can continue with stale data (or not be triggered at all).
To solve this issue, we can collect the directories that are searchable
via FILESEXTRAPATHS and add them to the do_kernel_metadata task
checksum.
This allows modifications, additions and removals from the potential
kernel feature directories to trigger a re-execution of the meta data
task.
Signed-off-by: Bruce Ashfield <bruce.ashfield@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
...instead we call read_subpackage_metadata.
Calling package_get_auto_pr *should* result in the same PKGV AUTOINC
replacement. However, it will also end up changing PKGR differently
then do_package as the BB_TASKHASH used will be for the wrong task.
Generally this won't cause any real-world issue, but it could cause
problems.
Moving to read_subpackage_metadata ensures that the values used
in do_package will be read in and used for kernel deployment.
Signed-off-by: Mark Hatle <mark.hatle@kernel.crashing.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Using this mechanism ensures that we have a single point to implement
the loading of the package and subpackage meta data. This also then
allows the buildhistory class to use the regular datastore vs it's
own custom arrays for processing history items.
Signed-off-by: Mark Hatle <mark.hatle@kernel.crashing.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
It was requested to add the ability to upgrade configuration warnings
to errors, so that they can't be missed in a build.
Add a flag KMETA_AUDIT_WERROR, that when set, triggers a bb.fatal at
the end of configuration checking if any warnings are found.
This is off by default, but is available for those that want to enable
it in their kernel recipe or bbappend.
Signed-off-by: Bruce Ashfield <bruce.ashfield@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
provides
emit_pkgdata() creates symlinks for each of the RPROVIDES in
pkgdata/MACHINE/runtime-rprovides. However this string can contain
versions which results in directories called (=2.32), so pass the
RPROVIDES string through bb.utils.explode_deps() to strip the versions
out.
Helps mitigate - but not solve - #13999.
Signed-off-by: Ross Burton <ross.burton@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Add ASSUME_PROVIDED to WICVARS so that wic can use that
when running on bitbake environment.
This fixes the below build failure in do_image_wic
when --include-path is used in wks file.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Prior work has refactored the do_install task multiple times, and any
references to PKGV and PKGR (even indirect ones) have been removed.
Signed-off-by: Mark Hatle <mark.hatle@kernel.crashing.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Sync up the anonymous python definition with the other package_*.bbclass
files. This should make future maintenance easier, even though it has
no difference in behavior from what was there.
Additional, there was a missing deltask in the nopackages.bbclass related
to the package_tar which has been corrected. This could cause problems on
native recipes when package_tar was enabled.
Signed-off-by: Mark Hatle <mark.hatle@kernel.crashing.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
There's a race between do_symlink_kernsrc and do_populate_lic, since
the latter is ordered "after do_patch"; so the two may run in
parallel. In some cases, that actually causes do_populate_lic to fail
if it happens to look for a license file somewhere under ${S} in the
short window after shutil.move and before the symlink has been
created.
Fix that by simply ordering symlink_kernsrc before do_patch. Any task
that pokes around in ${S} looking for files should be ordered after
do_patch, so this should also fix similar latent races with other ad
hoc tasks.
Signed-off-by: Rasmus Villemoes <rasmus.villemoes@prevas.dk>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
By doing this we can revert b18c32ab6bc9c4f1953e9f79aa39bc92d1c4e30d which
was a pretty ugly hack anyway and now means the different providers are all
being handled consistently.
Anyone with SDK recipes will need to ensure nativesdk-sdk-provides-dummy
is included in those builds (or an equivalent). This is a good thing to
do anyway.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
The output was unsorted, so different versions of python, different input
ordering could have have changed the files, and thus changed the hashes
making the system think the output was different, even when unmodified.
Signed-off-by: Mark Hatle <mark.hatle@kernel.crashing.org>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This class had great ideas and potential but now we have hash
equivalence and reproducibility, its effectively obsolete.
I'm not aware of any serious use of the class, we certainly don't
get bug reports or patches so remove it and focus on those other
areas.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
do_kernel_metadata gathers and sanitizes the meta-data that is used
in later steps of the kernel-yocto build process. The processing
takes the form of configuration and patch gathering.
The current single pass of both config and patch gathering means
that patches can't modify in-tree configuration elements (although
this makes tracing configuration changes harder, it is a valid
workflow).
We can divide the routine into a patch and configuration phase, and
call the config variant after patching is done. This keeps the
common parts of the gahering intact, but allows us flexibilty in
when the stages happen.
Signed-off-by: Bruce Ashfield <bruce.ashfield@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
To aid classifying options and streamling reports, fragments
and classification info are consolidated into a kernel meta
directory.
The name of the meta directory is (somewhat) flexible, and hence
is queried by the tools and created if it doesn't exist.
The recent enhancements to the .config auditing introduced a
new query for the meta directory, but isn't being run in the
kernel's source directory .. and is hence creating an unwanted
file in $TOPDIR. We set the kernel source directory as the cwd
for the call, and everything goes back to where it belongs.
Signed-off-by: Bruce Ashfield <bruce.ashfield@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
When compiling xilinx-zynq board linux-kernel-dev(v5.8) if
"GCC_PLUGINS=y", The following error will appear:
"HOSTCXX -fPIC scripts/gcc-plugins/arm_ssp_per_task_plugin.o
fatal error: gmp.h: No such file or directory"
the GCC_PLUGINS depend on return result of gcc-plugin.sh execution
however in gcc-plugin.sh use HOSTCC to detect the feature of GNU
extension of gcc, this will result that HOSTCC can compile the file
successfully, but HOSTCXX is used in the actual compilation process.
Signed-off-by: Zhang Qiang <qiang.zhang@windriver.com>
Signed-off-by: Bruce Ashfield <bruce.ashfield@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
CMake projects can specify the NONE project type. Projects that do this
do not use any C or C++ compiler, this currently works fine with caveat
that when changing the machine/arch the compiler is different causing
signature hash differences.
To avoid the signature hash differences clear the associated C/CXX
compiler variables. In order to achieve this with overrides, simplify
the existing construction of the values using a python function and
variable setting and remove the anonymous variable setup.
Signed-off-by: Nathan Rossi <nathan@nathanrossi.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Currently uninative handles SOURCE_MIRROR_URL but not generic PREMIRRORS.
It can handle this better, attempt to iterate PREMIRRORS entries.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
buildtools is built as a nativesdk which needs to use OECORE_NATIVE_SYSROOT
for its own purposes and can reset it. Save and restore the value
within the eSDK so the two don't clash.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Enable the kernel-yocto bbclass to use enhanced capabilities from
the kern-tools symbol_why.pl.
We bump the kern-tools SRCREV to pickup the reworking of symbol_why,
which uses Kconfiglib to provide analysis on configuration values.
This is useful for debugging why a symbol specified in a fragment
did not end up in the final .config.
We introduce two ways to interact with the new symbol_why:
1) a replacement of the existing kconf_check script
2) a dedicated task that is explicitly invoked to dump details
on the configuration.
The kconf_check replacement is transparent to the user, and is
run in exactly the same way as it was previously. But we get better
output and more detailed diagnostics if there are symbols that
don't make it into the final .config
The second way to interact with symbol why is via the new task
do_config_analysis. This is invoked like any other task, and by
default will provide a full configuration analysis and point the
user at files to look at for details.
If a more targetted analysis is desired, then specific symbols
can be set in the CONFIG_ANALYSIS variable. When this variable
is set, the task will only run for the given symbols and provide
per-variable links to the user. This variable can be set like
any other, including specification in the local.conf:
CONFIG_ANALYSIS_pn-linux-yocto-dev = 'NF_CONNTRACK LOCALVERSION'
Which produces output as follows:
WARNING: linux-yocto-dev-5.8-rc++gitAUTOINC+d22beb8f8a_8fc484ed37-r0
do_config_analysis: Configuration analysis executed, see: tmp/work/qemuarm64-poky-linux/linux-yocto-dev/5.8-rc++gitAUTOINC+d22beb8f8a_8fc484ed37-r0/NF_CONNTRACK-config-analysis.txt for details
WARNING: linux-yocto-dev-5.8-rc++gitAUTOINC+d22beb8f8a_8fc484ed37-r0
do_config_analysis: Configuration audit executed, see: tmp/work/qemuarm64-poky-linux/linux-yocto-dev/5.8-rc++gitAUTOINC+d22beb8f8a_8fc484ed37-r0/NF_CONNTRACK-config-audit.txt for details
WARNING: linux-yocto-dev-5.8-rc++gitAUTOINC+d22beb8f8a_8fc484ed37-r0
do_config_analysis: Configuration analysis executed, see: tmp/work/qemuarm64-poky-linux/linux-yocto-dev/5.8-rc++gitAUTOINC+d22beb8f8a_8fc484ed37-r0/LOCALVERSION-config-analysis.txt for details
WARNING: linux-yocto-dev-5.8-rc++gitAUTOINC+d22beb8f8a_8fc484ed37-r0
do_config_analysis: Configuration audit executed, see: work/qemuarm64-poky-linux/linux-yocto-dev/5.8-rc++gitAUTOINC+d22beb8f8a_8fc484ed37-r0/LOCALVERSION-config-audit.txt for details
Signed-off-by: Bruce Ashfield <bruce.ashfield@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Introduce the KCONFIG_CONFIG_ROOTDIR variable to allow recipes
which the .config file is not under the ${B} to select the root
directory location for the config file.
Signed-off-by: Diego Sueiro <diego.sueiro@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This patch adds support for adding configuration node even
when dtb is not part of the FIT image. The conf and default
node number are therefore changed to point to kernel ID rather
than dtb ID when dtb does not exist.
Signed-off-by: Usama Arif <usama.arif@arm.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
Fix a build-time race condition that resulted in intermittent build
failures in the do_assemble_fitimage task. The race condition involved
the do_assemble_fitimage task reading the vmlinux file while the
do_compile_kernelmodules task was re-writing the vmlinux file.
This can be seen with an aarch64 image build that uses a 5.4 based
kernel and sets KERNEL_DEVICETREE. The problem is that the do_compile
snippet that the kernel-devicetree class appends did not specify the
full kernel build environment when building the DTB(s) from the kernel
tree. This resulted in CONFIG_CC_CAN_LINK=y being removed from the
kernel config file just before the do_compile task completed. The
CONFIG_CC_CAN_LINK=y line was then re-inserted into the kernel config
file as part of the do_compile_kernelmodules task.
In some cases, this resulted in the do_compile_kernelmodules task to
re-link vmlinux which sometimes occured at the same time that the
do_assemble_fitimage task was attempting to use vmlinux. The
do_assemble_fitimage task would fail with the following error message:
aarch64-poky-linux-objcopy:vmlinux: file format not recognized
We can use the pine-a64-lts machine, from the meta-pine64 layer, to show
that the kernel config file was changed between do_compile and
do_compile_kernelmodules:
$ C=tmp/work/pine_a64_lts-poky-linux/linux-pine64/5.7+gitAUTOINC+ae03bade3b-r0/linux-pine_a64_lts-standard-build/.config
$ bitbake -c do_kernel_configcheck virtual/kernel
...
$ md5sum $C; grep CC_CAN_LINK $C
32b133cf8a749a91f698a7ca8616c84f ...
CONFIG_CC_CAN_LINK=y
$ bitbake -c do_compile virtual/kernel
...
$ md5sum $C; grep CC_CAN_LINK $C
2fd2ec2a66eecc329dcb5afaf005eada ...
$ bitbake -c do_compile_kernelmodules virtual/kernel
...
$ md5sum $C; grep CC_CAN_LINK $C
32b133cf8a749a91f698a7ca8616c84f ...
CONFIG_CC_CAN_LINK=y
With this change, the do_compile snippet appended by the
kernel-devicetree class does not modify the kernel config. The kernel
config is unchanged across the do_compile and do_compile_kernelmodules
tasks and do_compile_kernelmodules will not attempt to re-link vmlinux.
Signed-off-by: Tyler Hicks <tyhicks@linux.microsoft.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
KERNEL_FEATURES are explicitly checked by the kern-tools, and trigger
and error when not present. This is since the KERNEL_FEATURES are
specified by layers or BSPs as elements that are requied for correct
operation.
That being said, in order to support more kernel recipes that are using
fragments, but not necessarily the yocto kernel-cache for meta-data,
this change introduces: KERNEL_DANGLING_FEATURES_WARN_ONLY
As you'd expect, when set, missing kernel features only print a
message and warn that runtime issues may occur, but otherwise allow
the kernel configuration and build process to continue.
This was noticed/required when meta-virtualization adopted a more
generic kernel bbappend (to serve the needs of more layers in the
ecosystem), but also specifies some KERNEL_FEATURES. If layers using
a recipe derived from kernel-yocto, but not the main kernel-cache,
used meta-virtualization and error would be thrown due to missing
features. We now can both allow them to continue, and also provide
a more useful message to resolve the issue.
Signed-off-by: Bruce Ashfield <bruce.ashfield@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
|
This is required by newer meson versions in cross builds, as
acknowledgement that the tool in PATH is ok to use.
Signed-off-by: Alexander Kanavin <alex.kanavin@gmail.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
|
Peter A. Bigot
Rasmus Villemoes
Norman Stetter
Bruce Ashfield
Khem Raj
Konrad Weihmann
Jens Rehsack
Martin Jansa
Andrey Zhizhikin
Khairul Rohaizzat Jamaluddin
Ross Burton
Diego Sueiro
Chris Laplante
Matt Madison
Mark Hatle
Yanfei Xu
Sumit Garg
Ross Burton
Vijai Kumar K
Richard Purdie
Zhang Qiang
Nathan Rossi
Usama Arif
Tyler Hicks
Alexander Kanavin