| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
In previously patch, I created new guest image for ivi demo.
As a trial result using initial patch, I think so need to common
OVERRIDES keyword and specific OVERRIDES keyword.
This patch create cluster specific config and modify OVERRIDES keyword.
Bug-AGL: SPEC-4099
Signed-off-by: Naoto Yamaguchi <naoto.yamaguchi@aisin.co.jp>
Change-Id: I8c39f2a1fade2e0cc920a2efe9625678b07fb37b
|
|
AGL IC integration will have two guests, one of the cluster demo,
one of the simple ivi demo.
This patch add initial support for ivi container for demo.
Bug-AGL: SPEC-4099
Signed-off-by: Naoto Yamaguchi <naoto.yamaguchi@aisin.co.jp>
Change-Id: I4a1da6e30aee99e8ce2e2c404da78a413c647fd2
|
|
This is mostly the result of running a slightly customized version
of the convert-overrides.py script from poky with additional
overrides added. A few minor fixups were done by hand afterwards
during a review of the changes.
The intent of these changes is to minimize the effort to keep the
"next" branch that builds against poky master up to date and tested
in preparation for the switch to the next Yocto LTS release in
early 2022.
Bug-AGL: SPEC-4052
Signed-off-by: Scott Murray <scott.murray@konsulko.com>
Change-Id: I0bbae97331abee1367b368ce54ab39e3162a12ec
|
|
Note that these changes do allow building and booting the
lxc-host-image-demo image, and the cluster-demo container does
get autostarted, but currently Weston fails to start in the
container.
Changes:
- Add a meta-rcar-gen3 dynamic layer with some bbappends to
disable problematic kernel module dependencies when building for
the container guest configuration. The ones used are based on
the ones in the IC EG container-bsp github repo, but fewer are
needed due to the use of linux-dummy, and the changes are done
via aglcontainerguest overrides.
- Add linux-renesas bbappend that disables the problematic default
addition of virtio.scc to KERNEL_FEATURES done in the
meta-virtualization layer, which currently is dependent on the
file coming from the linux-yocto kernel metadata, which is not
the case with linux-renesas. Upstream have been engaged on this
being a problem, so this may not be required in the future.
- Add lxc.scc to KERNEL_FEATURES in linux-renesas bbappend to
leverage the .cfg from meta-virtualization.
- Borrow a couple of tweaks from the IC EG container-bsp github
repo to disable firmware and disable the serial consoles in the
container guest build.
Bug-AGL: SPEC-3953
Signed-off-by: Scott Murray <scott.murray@konsulko.com>
Change-Id: I30fe6a940c5c44ee2ad71410a97a4495829c0b45
|
|
Add meta-agl-lxc layer with LXC container integration recipes, and
"agl-lxc" feature to enable its use. The changes are based on the
previous prototypes by Tadao Tanikawa and the updates done for the
trial integration layers on the agl-ic-eg github by Naoto Yamaguchi,
Hiroyuki Ishii, and Jan-Simon Moeller.
The layer can be used by running aglsetup.sh with just the "agl-lxc"
feature, then building lxc-host-image-demo with bitbake.
Notes:
- The agl-lxc feature enables a "agl-container-guest" multiconfig
configuration for building the guest image(s). This suffices for
demonstrating integrated builds with multiconfig, but may change
in the future with extension of aglsetup.sh to add more
flexibility.
- There are guest-image-minimal and guest-image-cluster-demo recipes
that are intended as a base container template and a demo of its
use. The cluster-demo image contains the demo cluster reference
UI application, and is roughly equivalent to what is named
cluster-guest in the agl-ic-eg trial integration configuration.
- There are likewise lxc-host-image-minimal and lxc-host-image-demo
images. The host demo is configured to pull in the cluster-demo
guest image, and configuration has been added such that it will
automatically boot and start the cluster reference UI application.
- Unlike previous work where the guest container image is added to
the host image via a recipe that packages up the guest image's
contents, a more direct scheme is used where the guest image is
unpacked inside the host image rootfs via the use of the
ROOTFS_POSTPROCESS_COMMAND mechanism. This avoids some hassles
around potentially mismatched user and group id lists in the host
and guest images, with the additional benefit of being faster.
However, using this scheme is to some degree only necessary due to
the use of LXC with unpacked guest image hierarchies. With other
container tooling that uses e.g. OCI images, a revert to using
packages may be preferable.
- The reusable logic in the lxc-host-image-minimal recipe makes some
assumptions wrt naming. They are:
- guest image names have the format guest-image-<name>
- guest images have a corresponding LXC configuration file recipe
with the name lxc-config-<name>
The intent is to simplify reuse of the base minimal recipes, but
this can be reworked if deemed too restrictive.
- packagegroups for e.g. pipewire and weston have purposefully not
been defined at this time. It seems better to wait until it is
more obvious how useful they would be.
Limitations:
- The host image is currently rebuilt every time to work around an
issue with multiconfig dependencies that is being investigated
upstream. This can hopefully be removed in the near future.
- The demo images are currently in meta-agl-lxc for convenience.
It may make more sense for them to reside in another location to
ease shifting some things into meta-agl.
- At the moment, only qemux86-64 has been tested. The changes to
enable host and guest configuration tweaks for the rcar3 BSP need
further investigation and discussion on where they should be
maintained.
Bug-AGL: SPEC-3953, SPEC-3955
Signed-off-by: Scott Murray <scott.murray@konsulko.com>
Change-Id: I69d41f1d1c736f05cf3b4997dbda147879e2e25d
|
Naoto Yamaguchi
Scott Murray