xen-troops / meta-xt-images Goto Github PK

XEN is not deployed if yocto (lower one) is build with another build directory.

The way build artifacts are installed using Xen makefile and Yocto build differ, which makes it not so convenient to use SDK and daily build.
Recipe should be adjusted to match make install from Xen.

Commit acfb476 "gles-user-module: Install user files to canonical locations" broke build of rcar-proprietary-graphic target.
It fails on do_install() with message

install: cannot stat '/mnt/data1t/work_devu/build/build/tmp/work/x86_64-xt-linux/domd-image-weston/1.0-r0/repo/build/tmp/work/salvator_x_h3_4x2g_xt-poky-linux/rcar-proprietary-graphic/1.11-r0/xt-rcar//usr/bin/dlcsrv_REL': No such file or directory

Reason is that dlcsrv_REL is located in /usr/local/bin together with a lot of other tools.
And we can see in pvrinit and pvrlogdump scripts that /usr/local/bin/ is hardcoded path to some tools and dlcsrv_REL among them.

So we need to find solution that will take into account different ways used in different products.

Currently when we build host name is the same for all domains, so when you connect it is hard to tell which one it is.
It would very handy if we can set "/etc/hostname" differently for different domains

Please see #149 for details.

snb_be built for PulseAudio breaks the daily build

During lower yocto build there is a QA warning:

NOTE: recipe kmscube-git-r0: task do_package_qa: Started
WARNING: kmscube-git-r0 do_package_qa: QA Issue: kmscube rdepends on libgbm, but it isn't a build dependency, missing libgbm in DEPENDS or PACKAGECONFIG? [build-deps]

This causes a race and may lead to a build crash if gbm package is not installed to sysroot before yocto starts building kmscube.

There are recipes living in machine/meta-rcar-gen3 which are neither machine nor platform specific, e.g. libxenbe etc.
Move those out of RCar, so other platforms can also re-use.

meta-xt-images/recipes-domx/meta-xt-images-domx/recipes-graphics/wayland/weston_1.11.0.bbappend
is a product specific append and must be inside the product which uses it.
Please move to prod-devel, but also consult @aanisov if this is really needed

As we build Android as a recipe, not image it is then not possible to populate SDK. This needs to be fixed, so native environment used for the build can be re-used during development

It is unconfirmed, but seems that if (re-)built with sstate-cache enabled then llvm_build_dir gets installed into the root filesystem which is not correct.

Patches for bitbake fetch2 should be re-implemented while migrating to Yocto-thud in order to codebase changes in repo.py. The reason is xen-troops/manifests#69

This is a reminder task to check if it can be resolved with recent meta-openembedded releases.

Currently Android build requires that multilib is installed on the build host as according to this bug it is not possible to build multlib-native as of now.

Please integrate snd_be

During the build Android uses ccache which can be re-used for development purposes.
This needs to be provided in the way SSTATE cache is.

Move Linux version definition to corresponding product's recipe.

At least meta-virtualization and meta-selinux are odd for DomU image build.