Khem Raj [Thu, 14 May 2020 05:03:19 +0000 (22:03 -0700)]
gcc: Do not set -march for arm64 for libatomic
libatomic has mind of its own when it comes to setting -march for arm64
which conflicts with -mcpu option we pass from environment in some cases
since we always pass -march/-mcpu in OE, its safe to remove this option
mcpu removal from cortex-a55 is no longer needed since the option
conflict is now removed from libatomic instead
Signed-off-by: Khem Raj <raj.khem@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
If reproducible binaries are requested, generate the kernel modules
tarball in a reproducible way. Namely, sort the tarball contents, clamp
the mtime to the source date epoch, compress the archive with a
consistent compression level (9), and don't include the name and
timestamp in the compressed file.
[YOCTO #12719]
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Khem Raj [Tue, 12 May 2020 18:30:52 +0000 (11:30 -0700)]
gcc10: Default back to -fcommon
For time being change back to -fcommon as default, helps us iron
out other issues, eventually this should be removed as we fix the
packages to work with -fno-common
Signed-off-by: Khem Raj <raj.khem@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Khem Raj [Tue, 12 May 2020 18:30:51 +0000 (11:30 -0700)]
gcc10: Update to GCC 10.1 Release
* Package new gomp header acc_prof.h
* Package lto-dump which is a new tool in gcc10
* All Changes are here [1]
* Porting apps to gcc 10 help is here [2]
* Backport a patch to fix CET errors on cross builds
* Add patch to fix mingw libstdc++
populate_sdk_base.bbclass: fix warning: name not matched
When do populate_sdk with SDK_ARCHIVE_TYPE = "zip", have below warning
in log file, this patch fixes it.
zip warning: name not matched: sysroots/core2-64-poky-linux/etc/tmpfiles.d/etc.conf
zip warning: name not matched: sysroots/core2-64-poky-linux/etc/tmpfiles.d/home.conf
zip warning: name not matched: sysroots/core2-64-poky-linux/etc/resolv-conf.systemd
zip warning: name not matched: sysroots/core2-64-poky-linux/etc/mtab
zip warning: name not matched: sysroots/core2-64-poky-linux/etc/resolv.conf
zip warning: name not matched: sysroots/core2-64-poky-linux/var/lock
zip warning: name not matched: sysroots/core2-64-poky-linux/usr/lib/environment.d/99-environment.conf
zip warning: name not matched: sysroots/core2-64-poky-linux/sbin/klogd
zip warning: name not matched: sysroots/core2-64-poky-linux/sbin/syslogd
zip warning: name not matched: sysroots/x86_64-pokysdk-linux/usr/bin/chfn
zip warning: name not matched: sysroots/x86_64-pokysdk-linux/usr/bin/chsh
zip warning: name not matched: sysroots/x86_64-pokysdk-linux/usr/bin/passwd
zip warning: name not matched: sysroots/x86_64-pokysdk-linux/usr/bin/chpasswd
zip warning: name not matched: sysroots/x86_64-pokysdk-linux/sbin/vipw
zip warning: name not matched: sysroots/x86_64-pokysdk-linux/sbin/vigr
zip warning: name not matched: sysroots/x86_64-pokysdk-linux/sbin/nologin
Signed-off-by: Wenlin Kang <wenlin.kang@windriver.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Konrad Weihmann [Mon, 11 May 2020 19:28:06 +0000 (21:28 +0200)]
file: add PACKAGECONFIG for auto options
A few options of file configure are set to auto, which can lead to
unpredictable effects when something in the sysroot does provide
things that satisfy the autotools checks.
In the worst case this will lead to package-qa failures as libraries are
not set in RDEPENDS but configured for the tool.
To mitigate changes of accidental configure set explicit options via
newly introduced PACKAGECONFIG variables for bzip, lzma and zlib
support, where the default is just zlib, as it was before
Signed-off-by: Konrad Weihmann <kweihmann@outlook.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Yeoh Ee Peng [Fri, 1 May 2020 08:30:39 +0000 (16:30 +0800)]
selftest/imagefeatures: Enable sanity test for IMAGE_GEN_DEBUGFS
Add new testcase to check IMAGE_GEN_DEBUGFS. Test makes
sure that debug filesystem is created accordingly. Test also check
for debug symbols for some packages as suggested by Ross Burton.
[YOCTO #10906]
Signed-off-by: Humberto Ibarra <humberto.ibarra.lopez@intel.com> Signed-off-by: Yeoh Ee Peng <ee.peng.yeoh@intel.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Khem Raj [Tue, 5 May 2020 23:59:47 +0000 (16:59 -0700)]
gcc-runtime: Avoid march conflicts with newer cortex-a55 CPUs
gcc-runtime/libatomic explicitly add -march=armv8-a+lse for all arch64
but cortex-a55 is armv8.2-a, which essentially conflicts, so let gcc
override it by not forcing the -mcpu option from TUNE_CCARGS
Signed-off-by: Khem Raj <raj.khem@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Leitner [Wed, 6 May 2020 07:24:30 +0000 (09:24 +0200)]
libtirpc: remove extra "-fcommon" from CFLAGS
commit daed7eedba ("Avoid multiple-definiton with gcc -fno-common") of
libtirpc fixed the -fno-common issues. Therefore it should be save to
remove the extra "-fcommon" in CFLAGS.
Signed-off-by: Richard Leitner <richard.leitner@skidata.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Signed-off-by: Anton Eliasson <anton.eliasson@axis.com> Reviewed-by: Ola x Nilsson <ola.x.nilsson@axis.com> Signed-off-by: Anton Eliasson <anton.eliasson@axis.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
runqemu: force software x11 renderer for SDL when GL is not in use
If libsdl is configured with opengl enabled, it will try to use GL
renderers to open a window even if qemu doesn't use virgl. Setting
the environment variable forces a software X11 renderer, which is the
same one used when libsdl has opengl switched off. This avoids
needlessly loading the GL libraries and drivers from the host.
Signed-off-by: Alexander Kanavin <alex.kanavin@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Konrad Weihmann [Sat, 9 May 2020 20:32:44 +0000 (22:32 +0200)]
pypi.bbclass: mind package suffix on version check
Some pypi packages do have suffixes like dev, or a0 or b1.
When doing a version check on these, the version will get falsely
identified as major release versions.
Add a terminating slash to rule out those false positives
Signed-off-by: Konrad Weihmann <kweihmann@outlook.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Joshua Watt [Tue, 5 May 2020 16:39:10 +0000 (11:39 -0500)]
bitbake.conf: Prevent git from detecting parent repo in recipe
Prevents git commands run in a recipe from moving up past ${WORKDIR}
when searching for a .git directory, and thus prevents them from
detecting the parent OE-core .git directory. Fixes several
reproducibility issues where recipes would use the OE-core version as
the recipe version due to git walking up the tree.
Signed-off-by: Joshua Watt <JPEWhacker@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
* %_prefer_color is used by rpm to determine which color's ELF file
is preferred to be installed.
Here are file colors:
0 is unknown or other
1 is Elf32
2 is Elf64
4 is MIPS64 n32 (this color is added by oe-core's patch)
if default value set to 7, all colors are preferred color, always
be last-in-wins.
For this scenario, when we have 64bits python3 installed first,
then install 32bits python3 later, 64bits python3 will be overwrited,
and sys.path will point to /usr/lib, not /usr/lib64, this may cause
some python3 modules not work. so fixed by remove setting of default
value 7, and use default value 2 of rpm
* other distro like fedora also use the default %_prefer_color 2
Signed-off-by: Changqing Li <changqing.li@windriver.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Changqing Li [Fri, 8 May 2020 02:30:32 +0000 (10:30 +0800)]
rpm: fix file conflicts for MIPS64 N32
The following error occurred when prefer_color set to 2:
Error: Transaction check error:
file /sbin/ldconfig conflicts between attempted installs of
ldconfig-2.31+git0+71f2b249a2-r0.mips64_n32 and
lib32-ldconfig-2.31+git0+71f2b249a2-r0.mips32r2
file /usr/bin/gencat conflicts between attempted installs of
lib32-libc6-utils-2.31+git0+71f2b249a2-r0.mips32r2
...
This was because:
transactions_color = 001 (ELF32) & 010 (ELF64) & 100 (ELF32 N32 MIPS64)
FColor = Current file color (001) & transaction_color (111)
oFcolor = Previous file color (100) & transaction_color (111)
when "neither preferred" happened, handled as conflicts. this is too
restrictive for three way conflicts(mips64/mips64 n32/mips(32)).
Fixed by perform a 'last-in-wins' resolution when "neither is preferred".
Mesa needs libva.pc and libva headers to enable the VAAPI
state tracker and drivers.
This recipe is a variant of the full libva package build as in:
* it only depends on libdrm to build so it doesn't introduce
the circular dependency between mesa and libva, and
* it doesn't include the libraries in the final package.
However, there is another issue with build dependency handling
in Yocto. libva depends on mesa and mesa depends on this package.
Any package that depends on libva therefore would pull in libva
and this package resulting in an error in the prepare-sysroot
phase because they would install identical files into the
per-recipe sysroot.
Using the package name "*-initial" avoids this because of the
interaction between sstate.bbclass and staging.bbclass: any
package with the pattern "*-initial" in the name is excluded
from the dependency list unless explicitly added to DEPENDS.
To enable the VAAPI state tracker and drivers in mesa, it needs
libva.pc and the libva headers. To enable GLX support in libva,
it needs mesa to be compiled first. At the recipe level, this
would create a circular dependency between libva and mesa.
This is a preparation step before introducing a new libva recipe
variant to break the circular dependency.
Yi Zhao [Thu, 7 May 2020 03:24:32 +0000 (11:24 +0800)]
opkg-keyrings: check if opkg-key exists before run postinst
By default, the opkg-key command is not included in pokg package because
it is only installed when gpg support is enabled. We'd better check if
it exists before run 'opkg-key populate' in pkg_postinst.
Signed-off-by: Yi Zhao <yi.zhao@windriver.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
When vpnc support is included through PACKAGECONFIG, there
is now an extra vpn-script coming after the atest upgrade,
include that script into FILES so it gets packaged.
Signed-off-by: Alejandro Hernandez Samaniego <alejandro@enedino.org> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Debian 9 has an older version of mesa, where dri drivers do not link
against glapi explicitly, which causes problems when they are loaded
by newer mesa-native:
runqemu: set host DRI driver path for all instances of virgl usage
Falling through to host GL stack via epoxy chrpath hack was
found to be problematic in case of SDL particularly (because it does
not actually use epoxy, and does its own dlopen()).
So let's just use the mesa-native libraries everywhere, but instruct
them to load DRI drivers from the host.
NVidia's proprietary blob users have to take care of themselves, I'm afraid.
Signed-off-by: Alexander Kanavin <alex.kanavin@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
I2C tools are built by default in busybox. OE-core on the other hand
disables them all in the busybox defconfig except i2ctransfer. Since
this looks like an oversight, this patch proposes to disable this tool
as well.
Signed-off-by: Bartosz Golaszewski <bgolaszewski@baylibre.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This was found to be necessary for libsdl-native to enable opengl, as otherwise
the opengl check would fall through to the host with unpredictable outcome.
Signed-off-by: Alexander Kanavin <alex.kanavin@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Richard Purdie [Wed, 6 May 2020 13:35:39 +0000 (14:35 +0100)]
insane: Promote warnings into errors
There are several warnings the build can emit which show real problems
in the way recipes are being built. Some distros like poky and the Yocto
Project autobuilder have been fixing the issues these show for some time.
OE-Core has therefore been clean of these errors and warnings for some time.
Promote warnings into errors to match the defaults in poky to encourage
people to fix these real issue and reduce confusion when people switch
distros or customise.
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This is new in 0.60, and needs to be added explicitly
to cause vte to rebuild properly against icu changes
(there will be a QA check to catch such situations).
Signed-off-by: Alexander Kanavin <alex.kanavin@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Andrej Valek [Fri, 1 May 2020 09:15:54 +0000 (11:15 +0200)]
icu: upgrade 66.1 -> 67.1
- 0001-icu-Added-armeb-support.patch - rebased
- 0001-Fix-big-endian-build.patch - removed, already included in new version
- CVE-2020-10531.patch - removed, already included in new version
- icu-pkgdata-large-cmd.patch - removed, implemented correct size
Signed-off-by: Andrej Valek <andrej.valek@siemens.com>
The tarball download website is full of broken links;
while direct tarball download still works, exploring
around and checking for latest versions does not.
Signed-off-by: Alexander Kanavin <alex.kanavin@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
testsdkext/devtool: initialize the test component's git repo
Devtool is relying on externalsrc class, which, in order to determine
if a rebuild is needed, relies on git to checksum files (if the
component tree is a git repo), or sets a flag to always rebuild if
the component tree is not a git repo.
This is problematic in testsdkext scenario, where the test component
is inside a build directory, which itself is inside the poky repo
checkout, and listed in .gitignore. What happens is that git walks
up the tree and uses the index of the poky repo. This works okay
with older versions of git, but git 2.26 complains that we're inside
a directory that is ignored, and returns an error.
To fix the issue, the git repository is initialized directly in the
component directory, just prior to running the tests.
Signed-off-by: Alexander Kanavin <alex.kanavin@gmail.com> Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
